The Finite Element Method using MATLAB Young WW. won Hyochoong Bang CRC PressCRC MECHANICAL ENGINEERING SERIES Edited by Frank A. Kulacki, University of Minnesota Published Entropy Generation Minimization Adrian Bejan, Duke University The Finite Element Method Using MATLAB Young W. Kwon, Naval Postgraduate Schoo! Hyochoong Bang, Korea Aerospace Research Institute To be Published Fundamentals of Environmental Discharge Modeling Lorin R. Davis, Oregon State University Mechanics of Composite Materials Autar K. Kaw, University of South Florida Nonlinear Analysis of Structures M. Sathyamoorthy, Clarkson University Mechanics of Solids and Shells Gerald Wempter, Georgia Institute of Technology Viscoelastic Solids Roderic Lakes, University of lowaLIMITED WARRANTY CRC Press warrants the physical diskette(s) enclosed herein to be free of defects in materials and workmanship for a period of thirty days from the date of purchase. If within the warranty period CRC Press receives written notification of defects in materials or workmanship, and such notification is determined by CRC Press to be correct, CRC Press will replace the defective diskette(s). The entire and exclusive liability and remedy for breach of this Limited Warranty shall be limited to replacement of defective diskette(s) and shall not include or extend to any claim for or right to cover any other damages, including but not limited to, loss of profit, data, or use of the software, or special, incidental, or consequential damages or other similar claims, even if CRC Press has been specifically advised of the possibility of such damages. In no event will the liability of CRC Press for any damages (o you or any other person ever exceed the lower suggested list price or actual price paid for the software, regardless of any form of the claim. CRC Press specifically disclaims all other warranties, express or implied, including but not limited to, any implied warranty of merchantability or fitness for a particular purpose. Specifically, CRC Press makes no representation or warranty that the software is fit for any particular purpose and any implied warranty of merchantability is limited to the thirty-day duration of the Limited Warranty covering the physical diskette(s) only (and not the software) and is otherwise expressly and specifically disclaimed, Since some states do not allow the exclusion of incidental or consequential damages, or the limitation on how long an implied warranty lasts, some of the above may not apply to you. DISCLAIMER OF WARRANTY AND LIMITS OF LIABILITY: The author(s) of this book have used their best efforts in preparing this material. These efforts include the development, research, and testing of the theories and programs to determine their effectiveness. Neither the author(s) nor the publisher make warranties of any kind, express or implied, with regard to these programs or the documentation contained in this book, including without limitation warranties of merchantability or fitness for a particular purpose. No liability is accepted in any event for any damages, including incidental or consequential damages, lost profits, costs of lost data or program material, or otherwise in connection with or arising out of the furnishing, performance, or use of the programs in this book. Library of Congress Cataloging-in-Publication Data Kwon, Young W. The finite element method using MATLAB / by Young W. Kwon and Hyochoong Bang. p. cm. — (The mechanical engineering series) Includes bibliographical references and index. ISBN 0-8493-9653-0 (alk. paper) 1. Finite clement method—Data processing. 2. MATLAB I. Bang, Hyochoong. IL Title IIL. Series: Advanced topics in mechanical engineering series. TA347.F5K86 1996 624.1'7'028553—de20 96-19091 cr This book contains information obtained from authentic and highly regarded sources. Reprinted material is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable efforts have been made to publish reliable data and information, but the author(s) and the publisher cannot assume responsibility for the validity of all materials or for the consequences of their use. Neither this book nor any part may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, microfilming, and recording, or by any information storage or retrieval system, ‘without prior permission in writing from the publisher. The consent of CRC Press LLC does not extend to copying for general distribution, for promotion, for creating new works, or for resale. Specific permission must be obtained in writing from CRC Press LLC for such copying. Direct all inquiries to CRC Press LLC, 2000 Corporate Blvd. N.W., Boca Raton, Florida 33431. © 1997 by CRC Press LLC No claim to original U.S. Goverment works Intemational Standard Book Number 0-8493-9653-0 Library of Congress Card Number 96-19091 Printed in the United States of America 34567890 Printed on acid-free paperTABLE OF CONTENTS CHAPTER 1. INTRODUCTION TO MATLAB Lt 12 1.3 L4 1.5 1.6 4 Hd 18 19 1.10 Ba 1.12 1.13 114 115 Finite Element Method Overview of the Book About MATLAB Vector and Matrix Manipulations Matrix Functions Data Analysis Functions Tools for Polynomials Making Complex Numbers Nonlinear Algebraic Equations Solving Differential Equations Loop and Logical Statement Writing Function Subroutines File Manipulation Basic Input-Output Functions Plotting Tools CHAPTER 2. APPROXIMATION TECHNIQUES 21 2.2 2.3 24 25 2.6 27 Methods of Weighted Residual Weak Formulation Piecewise Continuous Trial Function Galerkin’s Finite Element Formulation Variational Method Rayleigh-Ritz Method Rayleigh-Ritz Finite Element Method Problems CHAPTER 3. FINITE ELEMENT PROGRAMMING 3.1 3.2 3.3 34 3.5 CHAPTER 4. DIRECT APPROACH WITH SPRING SYSTEM 4l 42 4.3 Overall Program Structure Input Data Assembly of Element Matrices and Vectors Application of Constraints Example Programs Problems Linear Spring ‘Axial Member Torsional Member PRVVEE EEE RSSSRUSRESSanwnmee 31 34 37 42 42 43 46 51 52 54 69 31 51 7144 Other Systems Problems CHAPTER 5. LAPLACE’S AND POISSON’S EQUATIONS 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12 Governing Equation Linear Triangular Element Bilinear Rectangular Element Boundary Integral ‘Transient Analysis Time Integration Technique Axisymmetric Analysis ‘Three-Dimensional Analysis MATLAB Application to 2-D Steady State Analysis MATLAB Application to Axisymmetric Analysis MATLAB Application to Transient Analysis MATLAB Application to 3-D Steady State Analysis Problems CHAPTER 6. ISOPARAMETRIC ELEMENTS One-Dimensional Elements Quadrilateral Elements ‘Triangular Elements Gauss Quadrature MATLAB Application to Gauss Quadrature MATLAB Application to Laplace Equation Problems CHAPTER 7. TRUSS STRUCTURES ae 7.2 7.3 74 15 7.6 One-Dimensional Truss Plane Truss Space Truss MATLAB Application to Static Analysis MATLAB Application to Eigenvalue Analysis MATLAB Application to Transient Analysis Problems CHAPTER 8. BEAM AND FRAME STRUCTURES 8.1 8.2 8.3 8.4 8.5 Euler-Bernoulli Beam Timoshenko Beam Beam Elements With Only Displacement Degrees of Freedom Mixed Beam Element Hybrid Beam Element 17 80 83 86 90 92 96 101 103 107 119 124 148 153 157 160 168 170 176 183 193 197 199 203 204 215 223 231 235 242 245 249 253 83 157 197 2358.6 8.7 8.8 8.9 8.10 8.11 8.12 8.13 8.14 Composite Beams Two-Dimensional Frame Element Three-Dimensional Frame Element MATLAB Application to Static Analysis MATLAB Application to Eigenvalue Analysis MATLAB Application to Transient Analysis MATLAB Application to Modal Analysis of Undamped System MATLAB Application to Modal Analysis of Damped System MATLAB Application to Frequency Response Analysis Problems CHAPTER 9. ELASTICITY PROBLEM 91 9.2 9.3 9.4 9.5 9.6 9.7 9.8 9.9 Plane Stress and Plane Strain Force Vector Energy Method Three-Dimensional Solid Axisymmetric Solid Dynamic Analysis MATLAB Application to 2-D Stress Analysis MATLAB Application to Axisymmetric Analysis MATLAB Application to 3-D Stress Analysis Problems CHAPTER 10. PLATE STRUCTURES 10.1 10.2 10.3 10.4 10.5 10.6 10.7 Classical Plate Theory Classical Plate Bending Element Shear Deformable Plate Element Plate Element With Displacement Degrees of Freedom Mixed Plate Element Hybrid Plate Element MATLAB Application Problems CHAPTER 11. CONTROL OF FLEXIBLE STRUCTURES ILL 11.2 11.3 11.4 11.5 11.6 11.7 11.8 11.9 11.10 Introduction Stability Theory Stability of Multiple Degrees of Freedom Systems Analysis of a Second Order System State Space Form Description Transfer Function Analysis Control Law Design for State Space Systems Linear Quadratic Regulator Modal Control for Second Order Systems Dynamic Observer 256 259 264 265 280 284 288 296 299 304 307 311 313 315 318 320 321 334 345 356 361 364 366 368 373 378 382 391 393 394 398 401 406 417 427 438 448 451 307 393PREFACE The finite element method has become one of the most important and useful engineering tools for engineers and scientists. This book presents introductory and some advanced topics of the Finite Element Method (FEM). Finite element theories, formulations, and various example programs written in MATLAB! are presented. The book is written as a textbook for upper level undergraduate and lower level graduate courses, as well as a reference book for engineers and scientists who want to write quick finite element analysis programs. Understanding basic program structures of the Finite Element Analysis (FEA) is an important part for better comprehension of the finite element method. MATLAB is especially convenient to write and understand finite element analysis programs because a MATLAB program manipulates matrices and vectors with ease. These algebraic operations constitute major parts of the FEA program. In addition, MATLAB has built-in graphics features to help readers visualize the numerical results in two- and/or three-dimensional plots. Graphical presentation of numerical data is important to interpret the finite element results. Because of these benefits, many examples of finite element analysis programs are provided in MATLAB. The book contains extensive illustrative examples of finite element analyses using MATLAB program for most problems discussed in the book. Subroutines (MATLAB functions) are provided in the appendix and a computer diskette which contains all the subroutines and example problems is also provided. Chapter 1 has a brief summary of useful MATLAB commands which can be used in programming FEA. Readers may refer to MATLAB manuals for additional information. However, this chapter may be a good start for readers who have no experience with MATLAB. Subsequent chapters are presented in a logical order. Chapter 2 discusses the weighted residual method which is used for the formulation of FEA in the remaining chapters. Initially, continuous trial functions are used to obtain approximate solutions using the weighted residual method. Next, piecewise continuous functions are selected to achieve approximate solutions. Then, FEM is introduced from the concept of piecewise continuous functions. Finally, classical variational formulations are compared with the weighted residual formulations. Chapter 3 shows the basic program structure of FEA using ordinary differential equations for an one-dimensional system. MATLAB programs are provided to explain the programming. Both program input and output as well as internal program structure are fully discussed. A direct FEM approach using simple mechanics is presented in Chapter 4. This chapter: presents the basic concept of FEM using an intuitive and physical approach. MATLAB is a registered trademark of The MathWorks, Inc. For additional informationa contact: The MathWorks, Inc. 24 Prime Park Way Natick, MA 01760 phone: (508) 647-7000, fax: (508) 647-7001Finite element formulations for partial differential equations are presented in Chapter 5. This chapter explains not only domain integration for computation of the finite element matrices but also boundary integration to compute column vectors. Applications of Laplace’s equation to two- and three-dimensional domains as well as an axially symmetric domain are presented for both steady-state and transient problems. Chapter 6 shows concepts and programming of isoparametric finite elements. Because a complex shape of domain with curved boundary can be easily handled using isoparametric finite elements, these elements are very useful and common in FEA. Both one-dimensional and two-dimensional isoparametric elements are presented. A numerical technique and its programming concept are also discussed. As a program example, Laplace’s equation is solved using isoparametric elements. Chapters 7 and 8 discuss truss and frame structures. Static, dynamic, and eigenvalue problems are solved. In addition, one-, two- and three-dimensional structures are considered. As a result, coordinate transformation from local to global axes is explained. In particular, various formulations for the beam structure are compared; the relative advantages and disadvantages of each are cited. Modeling of laminated beams with embedded cracks is also discussed. Further, Chapter 8 presents the modal analysis and Fast Fourier Transform. Elasticity is studied im chapter 9. Plane stress/strain, axisymmetric and three dimensional problems are included. Both static and dynamic analyses are presented. The finite element formulations are presented in terms of the weighted residual method. However, an energy method is also discussed for comparison. Plate bending is given in Chapter 10. Similar to beam formulations, different plate bending formulations are presented for comparison. Finally, structural control using FEM is presented in Chapter 11. This chapter is intended to provide a broad understanding of the basic concepts of control law in conjunction with FEM. Due to limited space, only a few major control theories are presented. It is assumed that readers are already familiar with fundamentals of linear dynamic systems analysis. This book contains more material than can be covered in a one-semester. Thus, materials may be selected depending on course objectives. For an introductory FEM course, Chapters 2 through 9 are recommended. Depending on the desired course contents, some sections may be deleted. We would like to thank individuals who have contributed to this book. The authors would like to express our appreciation to Professor Aleksandra Vinogradov for reviewing the manuscript and providing us with many useful suggestions. We are also indebted to the staffs of CRC Press for their professional guidance in the production of this book. Finally but not lastly the authors sincerely appreciate the lifelong support and encouragement by their parents. YY. W. Kwon H. C. BangCHAPTER ONE INTRODUCTION TO MATLAB 1.1 Finite Element Method In order to analyze an engineering system, a mathematical model is developed to describe the system. While developing the mathematical model, some assumptions are made for simplification. Finally, the governing mathematical expression is developed to describe the behavior of the system. The mathematical expression usually consists of differential equations and given conditions. These differential equations are usually very difficult to obtain solutions which explain the behavior of the given engineering system. With the advent of high perfor- mance computers, it has become possible to solve such differential equations. Various numerical solution techniques have been developed and applied to solve numerous en- gineering problems in order to find their approximate solutions. Especially, the finite element method has been one of the major numerical solution techniques. One of the major advantages of the finite element method is that a general purpose computer program can be developed easily to analyze various kinds of problems. In particular, any complex shape of problem domain with prescribed conditions can be handled with ease using the finite element method. The finite element method requires division of the problem domain into many subdomains and each subdomain is called a finite element. Therefore, the problem domain consists of many finite element patches. 1.2 Overview of the Book ‘This book is written as a textbook for engineering students as well as a reference book for practicing engineers and researchers. The book consists of two parts: theory and program. Therefore, each chapter has initial sections explaining fundamental theories and formulations of the finite element method, and subsequent sections showing examples of finite element programs written in the MATLAB program. The collection of MATLAB function files (ie. m-files) used in the example programs is summarized in Appendix A and provided in a separate computer disc. 12 Introduction to MATLAB Chapter 1 A brief summary of some of MATLAB commands is provided in the following sections for readers who are not familiar with them. Those are the commands which may be used in finite element programs. Especially, the MATLAB commands for matrix operation and solution are most frequently used in the programs. For visual- ization of the finite element solution, some plotting commands are also explained. 1.3 About MATLAB MATLAB is an interactive software which has been used recently in various areas of engineering and scientific applications. It is not a computer language in the normal sense but it does most of the work of a computer language. Writing a computer code is not a straightforward job; typically boring and time consuming for beginners. One attractive aspect of MATLAB is that it is relatively easy to learn. It is written on an intuitive basis and it does not require in-depth knowledge on operational principle of computer programming like compiling and linking in most of other programming languages. This could be regarded as a disadvantage since it prevents users from understanding the basic principle in computer programming. The interactive mode of MATLAB may reduce computational speed in some applications. The power of MATLAB is represented by the length and simplicity of the code. For example, one page of MATLAB code may be equivalent to many pages of other computer language source codes. Numerical calculation in MATLAB uses collections of well written scientific/mathematical subroutines such as LINPACK and EISPACK. MATLAB provides Graphical User Interface (GUI) as well as three-dimensional graphical animation. In general, MATLAB is a useful tool for vector and matrix manipulations. Since the majority of the engineering systems are represented by matrix and vector equations, we can relieve our workload to a significant extent by using MATLAB. The finite element method is a well defined candidate for which MATLAB can be very useful as a solution tool. Matrix and vector manipulations are essential parts in the method. MATLAB provides help menu so that we can type help command when we need help to figure out a command. The help utility is quite convenient for both beginners and experts. 1.4 Vector and Matrix Manipulations Once we get into the MATLAB, we meet a prompt >> called MATLAB prompt. This prompt receives a user command and processes it providing the output on the next line. Let us try the following command to define a matrix. >> A= [1,3,6;2,7,8;0,3,9] Then the output appears in the next line as shown below. 136 A=2 7 8 039Section 1.4 Vector and Matrix Manipulations 3 Thus, a matrix is entered row by row, and each row is separated by the semi-colon(;). Within each row, elements are separated by a space or the comma(,). Commands and variables used in MATLAB are case-sensitive. That is, lower case letters are distinguished from upper case letters. The size of the matrix is checked with >> size(A) ans=3 3 ‘Transpose of a matrix In order to find the transpose of matrix A, we type >> A’ The result is 2 2 & ul orm Cn owe Column or row components MATLAB provides columnwise or rowwise operation of a matrix. The following expression >> A(:,3) yields 6 ans= 8 9 which is the third column of matrix A. In addition, >> A(1,:) represents the first row of A as ans=1 3 6 We can also try >> A(1,:) + A(3,:) as addition of the first and third rows of A with the result ans=1 6 15 Now let us introduce another matrix B as >> B = [8,4,5;6,7,2;8, 1,0]; ‘Then there seems to be no output on the screen. MATLAB does not prompt output on the screen when an operation ends with the semi-colon(;) at the end. If we want to check the B matrix again, we simply type >>B The screen output will be4 Introduction to MATLAB Chapter 1 wy " wae Ba oun Matrix addition Adding two matrices is straightforward like >>C=A+B at Al C=8 14 10 84 9 Thus we defined a new matrix C as sum of the previous two matrices. Matrix subtraction In order to subtract matrix B from matrix A, we type >>C=A-B 2 -1 1 Cc=-4 0 6 8 2 9 Note that C is now a new matrix not the summation of A and B any more. Matrix multiplication Similarly, matrix multiplication can be done as >>C=AxB 69 31 IL C= 112 65 24 90 30 6 1.5 Matrix Functions Manipulation of matrices is a key feature of the MATLAB functions. MATLAB is a useful tool for matrix and vector manipulations. Collections of representative MATLAB matrix functions are listed in Table 1.5.1. Examples and detailed explana- tions are provided for each function below. Matrix inverse The inverse of a matrix is as simple as >> inv(A) ans 1.8571 -0.4286 -—0.8571 —0.8571 0.4286 0.1905 0.2857 —0.1429 0.0476 In order to verify the answer, we can try >> Axinv(A);Section 1.5 Matrix Functions 5 Table 1.5.1 Basic Matrix Functions Symbol Explanations inv inverse of a matrix det determinant of a matrix rank rank of a matrix cond condition number of a matrix eye(n) the n by n identity matrix trace summation of diagonal elements of a matrix zeros(n,m) the n by m matrix consisting of all zeros which should be a 3 by 3 identity matrix. Determinant of a matrix >> d=det(A) produces the determinant of the matrix A. That is, d= 21 Rank of a matrix The rank of a matrix A, which is the number of independent rows or columns, is obtained from >>rank(A); Identity matrix >> eye(3) yields 100 ans=0 1 0 001 eye(n) produces an identity matrix of size n by n. This command is useful when we initialize a matrix. Matrix of random numbers A matrix consisting of random numbers can be generated using the following MATLAB function. >>rand(3, 3) 0.2190 0.6793 0.5194 ans = 0.0470 0.9347 0.8310 0.6789 0.3835 0.03466 Introduction to MATLAB Chapter 1 That is, rand(3,3) produces a 3 by 3 matrix whose elements consist of random numbers. The general usage is rand(n, m). trace Summation of diagonal elements of a matrix can be obtained using the trace operator. For example, >>Ca[l 39; 672 8 -1 -3; Then, trace(G) produces 6, which is the sum of diagonal elements of C. zero matrix >> zeros(5,4) produces a 5 by 4 matrix consisting of all zero elements. In general, zeros(n,m) is used for an n by m zero matrix. condition number The command cond(A) is used to calculate the condition number of a matrix A. The condition number represents the degree of singularity of a matrix. An identity matrix has a condition number of unity, and the condition number of a singular matrix is infinity. >>cond(eye(6)) ans = 1 An example matrix which is near singular is The condition number is >>cond(A) ans = 4.0000€+006 Further matrix functions are presented in Table 1.5.2. They do not include all matrix functions of the MATLAB, but represent only a part of the whole MATLAB functions. Readers can use the MATLAB Reference’s Guide or help command to check when they need more MATLAB functions. Matrix exponential The expm(A) produces the exponential of a matrix A. In other words, >> A=rand(3, 3) 0.2190 0.6793 0.5194 A= 0.0470 0.9347 0.8310 0.6789 0.3835 0.0346Section 1.5 Matrix Functions Table 1.5.2 Basic Matrix Functions (Continued) Explanations expm exponential of a matrix eig a rrr~—r—S lu LU decomposition of a matrix svd singular value decomposition of a matrix ar QR decomposition of a matrix \ used to solve a set of linear algebraic equations >>expm(A) 1.2448 0.0305 0.6196 ans = 1.0376 1.5116 1.3389 1.0157 0.1184 2.0652 Eigenvalues The eigenvalue problem of a matrix is defined as Ag = AG where 2 is the eigenvalue of matrix A, and ¢ is the associated eigenvector. >> e =eig(A) gives the eigenvalues of A, and >> [V, D] =eig(A) produces V matrix, whose columns are eigenvectors, and the diagonal matrix D whose values are eigenvalues of the matrix A. For example, >> A=[5 32% 146; 97 2 >> [V, D] =eig(A) 0.4127 0.5992 0.0459 V = 0.5557 -0.7773 —0.6388 0.7217 0.1918 0.7680 12.5361 0 0 eee 1.7486. 0 0 0 —3.2847 LU Decomposition The LU decomposition command is used to decompose a matrix into a combination of upper and lower triangular matrices, respectively. >>Aa[l 35 248 47 3); >> [L,U] =hu(A)8 Introduction to MATLAB Chapter 1 0.2500 1.0000 0 L= 0.5000 0.4000 1.0000 1.0000 0 0 4.0000 7.0000 3.0000 U= 0 1.2500 4.2500 0 0 4.8000 In order to check the result, we try >> LeU 135 ans=|2 4 8 473 The lower triangular matrix L is not perfectly triangular. There is another command available >> [L,U, P] =lu(A) 1.0000 0 0 L= 0.2500 1.0000 0 0.5000 0.4000 1.0000 4.0000 7.0000 3.0000 U= 0 1.2500 4.2500 0 4.8000 P= roo O° 1 0 0 Borneo Here, the matrix P is the permutation matrix such that P+ A= L+U. Singular value decomposition The svd command is used for singular value decomposition of a matrix. For a given matrix, A=UxV’ where ¥ is a diagonal matrix consisting of non-negative values. For example, we define a matrix D like >> D=[1 37; 29 5 28 5 The singular value decomposition of the matrix is >> [U, Sigma, V] =svd(D) which results in 0.4295 0.8998 +—-0.0775 U = | 0.6629 —0.3723 —0.6495 0.6133 ~0.2276 0.7564Section 1.5 Matrix Functions 9 15.6492 0 0 Sigma = 0 4.13330 0 0 0.1391 0.7771 —0.5982 —0.1956 0.1905 -0.0726 0.9790 ~ [0.5999 0.7980 —0.0576 QR decomposition A matrix can be also decomposed into a combination of an orthonormal matrix and an upper triangular matrix. In other words, A=QR where Q is the matrix with orthonormal columns, and 2 is the upper triangular matrix. The QR algorithm has wide applications in the analysis of matrices and associated linear systems. For example, 0.0470 0.9347 0.8310 0.2190 0.6793 0.5194 A= 0.6789 0.3835 0.0346 Application of the qr operator follows as >> [Q, R] =ar(A) yields —0.3063 —0.4667 —0.8297 Q= | -0.0658 —0.8591 0.5076 —0.9497 0.2101 0.2324 0 —1.0395 0.9490 —0.7149 —0.6338 —0.2466 R= 0 0 —0.0011 Solution of linear equations The solution of a linear system of equations is frequently needed in the finite element method. The typical form of a linear system of algebraic equations is written as Ar=y and the solution is obtained by >> # =inv(A) *y or we can use \ sign as >>e= Aly For example >>A=[1 3 4 5 7 8; 23 5)10 Introduction to MATLAB Chapter 1 Table 1.6.1 Data Analysis Functions Symbol Explanations min(max) minimum(maximum) of a vector sum sum of elements of a vector std standard deviation of a data collection sort sort the elements of a vector mean mean value of a vector used for componentwise operation of a vector and >> y= [10; 5 8); Let us compare two different approaches. >> [inv(A)*y A\y] 4.2500 —4.2500 ans= 1.7500 ‘1.7500 2.2500 2.2500 1.6 Data Analysis Functions Up to now, we discussed matrix related functions and operators of MATLAB. MATLAB has also data analysis functions for a vector or a column of a matrix. In Table 1.6.1, some operators for data manipulation are listed. Minimum (maximum) The min (max) finds a minimum (maximum) value of a given vector. For example, >>v=[ll 23 73 25 49 92 28 23); >>min(v) yields ans = PFE >>max(v) ans = 92 sum The sum command produces the summation of elements of a vector. For example,Section 1.6 Data Analysis Functions nay >> sum(v) yields ans = 924 Standard deviation The std command calculates the standard deviation of a vector. For example, >> std({l 4.10 —5 6 9 —20)) ans = 10.4517 Sort a vector The sort command is used to sort a vector in the ascending order. >> sort([l 4 10 -5 6 9 —20)) ans = 20-5 1469 10 Mean value of a vector The mean calculates the mean value of a vector. >> mean([1 4 10 -5 6 9 —20)) ans = 0.7143 Vector componentwise operation Let us define two vectors >> v1 = [1,5,6, 7]; v2 = (0, 2,3, 5); Sometimes we want to multiply components of v; with the corresponding components of v2. The operation is >> Ug = U1. * v2 ans = 0 10 18 35 In other words, ( ).*( ) represents the componentwise multiplication of two vectors. Another useful operator is >> v4 = 2./11 with vg=0 0.4 0.5 0.7143 Note that the data analysis tools explained in the above are applicable to matrices too. Each matrix column is regarded as a vector for data analysis.12 Introduction to MATLAB Chapter 1 Table 1.7.1 Polynomial Functions Symbol Explanations poly converts collection of roots into a polynomial equation roots finds the roots of a polynomial equation polyval evaluates a polynomial for a given value conv multiply two polynomials deconv | decompose a polynomial into a dividend and a residual polyfit curve fitting of a given polynomial 1.7 Tools for Polynomials Polynomials are frequently used in the analysis of linear systems. MATLAB provides some tools for handling polynomials. The summiary of polynomial functions is provided in Table 1.7.1. Roots of a polynomial equation A polynomial equation is given by ayn” age") 4--- ane + any) =0 ‘The roots of the polynomial equation is solved using roots command roots([a: a2 --- dn an4i]) For example, 24 +429 — 52? +62-9=0 >>roots([1 4 —5 6 —9]) yields ans = 5.2364 1.2008 0.0178 + 1.1963 0.0178 - 1.19631 Generation of a polynomial equation using roots The poly command takes the roots, and converts them into a polynomial equation. For instance, if we know [ru ray +++) te] in n=1 (2 —1ri)(@ — 2) ++ (@ = ta) = 2" + a2"? + age”? 4+ pant + dn thenSection 1.7 Tools for Polynomials 13 >>poly([ni, 72, +++, tal) provides us the coeflicients( (a1, a2, -++,n]) of the polynomial equation. For example, >>poly([-1 -—2+2*i —2-2*i -—5+7*i —5-7«i)) produces ans = 1 15 136 498 968 592 In order to check the result, we use roots command again. >>roots([1 15 136 498 968 592]) The result should be [-1 —24+2+i —2-2*i -5+7*i —5-7+i]. Polynomial value When we want to calculate the value of a polynomial at a certain point, we can use polyval. >> y=polyval({1 3 4 —5 ],2) ans = 23 which evaluates the polynomial s‘ + 367 + 4s —5 at s = 2. Multiplication of two polynomials The conv command is used to multiply two polynomials. For example, a(s)=s8?4+3s—1, 0(s) = 6? 287 +. 68-7 The multiplication of a(s) and 6(s) follows as >>c=conv([l 3 ~1],[1 -2 6 —7) ice 1 1 -1 13 -27 7 In other words, we obtain the coefficient vector c of the product of a(s) and 6(s). Decomposition of a polynomial The deconv is used to decompose a polynomial as a multiplicand and a residue. Let a(s) = b(s)m(s) + r(s) That is, the polynomial a(s) is represented in terms of a multiplicand m(s) and a residue r(s) via b(s). The MATLAB command is >> [m,r] =deconv(a, b) where the parameters are coefficient vectors for given polynomials. An example is given by >> [m,r] =deconv({1 —2 6 —7},{1 3 —1]) m=14 Introduction to MATLAB Chapter 1 = 0 0 22 -12 If we change the order of polynomials, >> [m,r] =deconv([l 3 -1],[1 -2 6 -7) m= 0 r= 13 -1 Polynomial fit The polyfit command is to generate a polynomial curve which fits a given set of data. The polynomial is obtained by minimizing the error between the polynomial and the given data set. The synopsis is p = polyfit(z, y,n) where # and y are vectors of the given data set in (2, y) form, and n is the order of the desired polynomial to fit the data set. The output result is p, the coefficient vector of the fitting polynomial. An example is provided below, >>ar=f[1 23 45 6); >>y=(-1 352 -3 I]; >> p =polyfit(z, y, 1) p= 2.1714 -1.2667 A linear curve fitting is performed for data set (2, y). 1.8 Making Complex Numbers In order to make a complex number 2+ 3 * i, we use >> 2+ 384i or >> 2+ 3aj MATLAB takes i and j as a pure complex number. In case i or j is defined already, we can use V/—1 as =sqrt(—1) i= 0+ 1.00001 >>Section 1.9 Nonlinear Algebraic Equations 15 Table 1.9.1 Functions for Nonlinear Algebraic Equations Symbol Explanations fmin finds minimum of a function of one variable fzero solves a nonlinear algebraic equation of one variable abs, angle For a given complex number, we use abs and angle commands to find out the magnitude (abs) and phase angle (angle) of the given complex number. For example, if >>pe=-l+i; then >> abs(c) ans = 1.4142 >> angle(c) ans = 2.3562 Real, imaginary parts of a complex number The real and imag are used to take the real and imaginary parts of a complex number. For example, >>e= 104987 >> [real(c),imag(c)] ans =—10 9 Conjugate The conj command is used to generate a complex conjugate number. For example >> conj(—1 +5 * i) ans = —1—54i 1.9 Nonlinear Algebraic Equations Nonlinear algebraic equations are frequently adopted in many different areas. ‘The nonlinear equations are different from linear equations, and there is no unique analysis tool to the nonlinear equations. MATLAB is equipped with some functions which can handle nonlinear equations. The list is presented in Table 1.9.1. Minimum of a function The MATLAB command fmin minimizes a function by finding out a value which minimizes the given function. The synopsis is fmin(' func’, 21,22)16 Introduction to MATLAB Chapter 1 Table 1.10.1 Numerical Techniques for Differential Equations Symbol Explanations ode23 solution using the 2nd/3rd order Runge-Kutta algorithm ode45 solution using the 4th/5th order Runge-Kutta algorithm where *func’ is the name of a function to be minimized and 21(z2) represents a lower(upper) limit of the interval of the function argument. For example, >>fmin('z + cos(2)’, -2,2) produces ans = —0.8603 Solution of a nonlinear algebraic equation When a nonlinear algebraic equation is written as 2° 43+ sin(z)— 25 =0 the MATLAB function fzero can be used to find a solution of the nonlinear algebraic equation. The synopsis is >> sol =fzero(' function’, 20) where ’function’ is a MATLAB function subroutine and 20 is an initial condition vector of the variables. For the given example, we write a function subroutine fetn.m as function [f] =fctn(z); f= 2°34 2 xcos(z) — 42; Then, we use fzero command as >> sol =fzero('fetn’,—5) sol = -2.1281 In order to check the solution >>fetn(—2.1281) ans = 1.91920-004 The error is due to the numerical format error. The number is truncated for screen display purpose, even if it is calculated using double precision format inside MATLAB. 1.10 Solving Differential Equations Linear and nonlinear differential equations can be also solved using MATLAB. List of numerical techniques solving differential equations is in Table 1.10.1.Section 1.11 Loop and Logical Statement 17 Runge-Kutta second and third order algorithm MATLAB uses the Runge- Kutta algorithm to solve a differential equation or a set of differential equations. The general synopsis is {t,2] = ode23(' func’, t0, tf, 20); where ’func’ is a function containing the derivative information, t0 (tf) is the initial (final) time, and 20 is an initial condition vector. The outputs are ¢, which contains the returned time points, and x which is the integrated output. For example, we want to solve # + sin(z) =0 which can be rewritten as ay = 22 2 = —sin(z1) where z; = 2 and 22 =. The ’func’ function should be provided as an independent function subroutine as func.m in a directory, which MATLAB can locate. Now we execute the ode23 command >> [t,2] =ode23(’ func’, 0, 10, 20); {3 is an initial condition and func.m is provided as function [f] =func(t,x); [f] = zeros(2,1); F(1) = 2(2); f(2) = —sin(x(1)); Runge-Kutta Fourth and Fifth order algorithm There is another Runge-Kutta algorithm ode45 which is more accurate than ode23. where [t, 2] = ode45(/ func’ ,t0,tf, 20) The same calling synopsis as ode23 can be applied to make use of the ode45 function. 1.11 Loop and Logical Statement There are some logical statements available in MATLAB which help us writing combinations of MATLAB commands. Furthermore, loop commands can be used as, in other programming languages. In fact, we can duplicate the majority of existing programs using MATLAB commands, which significantly reduces the size of the source18 Introduction to MATLAB Chapter 1 Table 1.11.1 Loop and Logical Statements Symbol Explanations for loop command similar to other languages while used for a Joop combined with conditional statement if produces a conditional statement elseif, else used in conjunction with if command break breaks a loop when a condition is satisfied codes. A collection of loop and logical statements in MATLAB is presented in Table 1.11.1. for loop The for is a loop command which ends with end command. >> for i=1:100 a(i,i) = 2* 4; end In the above example, jis a loop index which starts from 1 and ends at 100. There may be also multiple loops. >> for i=1:100 for j=1: for k= 1:50 a(é, f) = 8(,B) #o(b, 3) +a(4,3)5 end end end while The while command is useful for an infinite loop in conjunction with a conditional statement. The general synopsis for while command is as follows while condition statements end For example, i=l while (i < 100) i=i4]; end Another example of the while command is n= 1000; var = [];Section 1.11 Loop and Logical Statement 19 while (n > 0) n=n/2-1; var = var, n]; end The result is var = Columns 1 through 6 4.9900e+002 2.4850e+002 1.2325e+002 6.0625e+001 2.9313e+001 1.3656e+001 Columns 7 through 9 5.8281e+000 1.9141e+000 -4.2969e-002 where we used [ ] in order to declare an empty matrix. if, elseif, else The if, elseif, and else commands are conditional statements which are used in combination. if condition #1 statement #1 elseif condition #2 statement #2 else statement #3 end For example, n= 100; if (rem(n,3) == 0) 2=0; elseif (rem(n,3) == 1) z= else z end where rem(x,y) is used to calculate the remainder of z divided by y. break The break command is used to exit from a loop such as if and while. For example, for i= 1: 100 i=i+]; if(i == 10) break; end end20 Introduction to MATLAB Chapter 1 Table 1.11.2 Loop and Logical Statements Symbol Explanations two conditions are equal i two conditions are not equal <= (=) one is less (greater) than or equal to the other <(>) one is less (greater) than the other & and operator - two conditions are met not operator I or operator - either one condition is met Logical and relational operators The logical and relational operators of MATLAB are as listed in Table 1.11.2. The above command sets are used in combination. 1.12 Writing Function Subroutines MATLAB provides a convenient tool, by which we can write a program using collections of MATLAB commands. This approach is similar to other common programming languages. It is quite useful especially when we write a series of MATLAB commands in a text file. This text file is edited and saved for later use. The text file should have filename.m format normally called m-file. That is, all MATLAB subroutines should end with .m extension, so that MATLAB recognizes them as MATLAB compatible files. The general procedure is to make a text file using any text editor. If we generate a file called func/.m, then the file funci.m should start with the following file header function[ovy , ov, = funel(iv;,iv2,...) where iv, iv9,... are input variables while ov,0v2,... are output variables. The input variables are specific variables and the output variables are dummy variables, for which we can use any variables. For example, let us solve a second order algebraic equation. az? +br4+e=0 ‘The solution is given in analytical form as _ -b+t VP — Teac 2 QeaSection 1.12 Writing Function Subroutines 21 We want to write an m-file with the name secroot.m, which produces the analytical solution. function [r1,r2]=secroot(a, b,c); % % Find Determinant —— Any command in MATLAB which starts with % % sign is a comment statement Det =b2—-4+a%c; if (Det <0), r1 = (—b + j * sgrt(—Det))/2/a; r2= (-b — j * sqrt(—Det))/2/a; disp('The two roots are compler conjugates’); elseif( Det rl = —b/2/a; 72 = —b/2/a; disp('There are two repeated roots’); else(Det > 0) rl = (—b + sqrt(Det))/2/a; —b — sqrt(Det))/2/a; disp('The two roots are real’); end Some commands appearing in the above example will be discussed later. Once the secroot.m is created, we call that function as >> [r1,r2]=secroot(3, 4,5) or >> [pl, p2] =secroot(3, 4,5) One thing important about the function command is to set up the m-file pathname. The m-file should be in the directory which is set up by MATLAB configuration set up stage. In the recent version of MATLAB, the set up procedure is relatively easy by simply adding a directory which we want to access in a MATLAB configuration file. ‘Another function subroutine fet.m is provided below. function [f] =fct(z) fs(l-2/2; The above function represents f(z) = (1— 2). In the MATLAB command prompt, we call the function as >> y =fet(9); The function subroutine utility of MATLAB allows users to write their own subrou- tines. It provides flexibility of developing programs using MATLAB.22 Introduction to MATLAB Chapter 1 Table 1.13.1 File Manipulation Commands Symbol Explanations save save current, variables in a file load load a saved file into MATLAB environment, diary save screen display output in text format 1.13 File Manipulation Manipulating files is another attractive feature of MATLAB. We can save MATLAB workspace, that is, all variables used, in a binary file format and/or a text file format. The saved file can be also reloaded in case we need it later on. The list of file manipulation commands is presented in Table 1.13.1. save The save command is used to save variables when we are working in MATLAB. The synopsis is as follows save filename var; varz ... where filename is the filename and we want to save the variables, var;, vara, .... The filename generated by save command has extension of .mat, called a mai-file. If we do not include the variables name, then all current variables are saved automatically. In case we want to save the variables in a standard text format, we use save filename vary varz .../ascii/double load The load command is the counter-part of save. In other words, it reloads the variables in the file which was generated by save command, The synopsis is as follows load filename var, vary ... where filename is a mat-file saved by save command. Without the variables name specified, all variables are loaded. For example, >>a=([l 3 4]; >> b=3; >>save test >>clear all % clear all variables >>who % display current variables being used >>load test >>who diary Using diary command, we can capture all MATLAB texts including command and answer lines which are displayed on the screen. The texts will be saved in a file, so that we can edit the file later. For example, >>diary onSection 1.14 Basic Input/Output Functions 23 Table 1.14.1 Input-Output Functions Symbol Explanations input save current variables in a file disp load a saved file into MATLAB environment. format check the file status in the directory >pa=lb=4e >>[a be) >>d=axb >>e=geh >>diary off Now we can use any text editor to modify the diary file. The diary command is useful displaying the past work procedures. Also, it can be used to save data in a text format. 1.14 Basic Input-Output Functions Input/output functions in MATLAB provide MATLAB users a friendly pro- gramming environment. Some input/output functions are listed in Table 1.14.1. input The input command is used to receive a user input from the keyboard. Both numerical and string inputs are available. For example, >> age= input(/How Old are you?’) >> name =input('What is your name’, s') The ’s’ sign denotes the input type is string. disp The disp command displays a string of text or numerical values on the screen. It is useful when we write a function subroutine in a user-friendly manner. For example, >> disp(‘This is a MATLAB tutorial!) >> c=3%; >> disp('The computed value of c turns out to be’) >>e format The format command is used to display numbers in different formats. MATLAB calculates floating numbers in the double precision mode. We do not want to, in some situations, display the numbers in the double precision format on the screen. For a display purpose, MATLAB provides following different formats >>2=1/924 Introduction to MATLAB Chapter 1 Table 1.15.1 Plotting Commands Symbol Explanations plot basic plot command xlabel(ylabel) attach label to x(y) axis axis manually scale x and y axes text place a text on the specific position of graphic screen title place a graphic title on top of the graphic ginput produce a coordinate of a point on the graphic screen gtext receives a text from mouse input grid add a grid mark to the graphic window pause hold graphic screen until keyboard is hit subplot breaks a graphic window into multiple windows x=0.10111 >>format short e w= 1,111le-001 >>format long O.10111111111111 >>format long e @ = 1.111111111111111e — 001 >>format hex z = 3fbeTleTle71e7le 1.15 Plotting Tools MATLAB supports some plotting tools, by which we can display the data in a desired format. The plotting in MATLAB is relatively easy with various options available. The collection of plotting commands is listed in Table 1.15.1. ‘A sample plotting command is shown below. >>t=0:0.1: 10; >> y= sin(t); >> plot(y) >> title(’plot(y)’)Section 1.15 Plotting Tools 25, The resultant plot is presented at the top of Fig. 1.15.1. Soe oe >> y= sin(t); >> plot(t, y) >> title(’plot(t,y)’) The resultant plot is presented at the bottom of Fig. 1.15.1. In the above example, = 0: 0.1: 10 represents a vector ¢ which starts from 0 and ends at 10 with an interval of 0.1. We can use just y or both y and ¢ together. In the first case, the horizontal axis represents number of data, from 0 to 101. In the second case, the horizontal axis is the actual time scale ¢ in the plot(t,y) command. Plotting multiple data We plot multiple data sets as shown below. >>t=0:1: 100; >> yl = sin(t).*t; >> y2 = cos(t). +t; >> plot(t,y/,-4t,y2,’-)) where ’-’ and -’ represent line styles. The line styles, line marks, and colors are listed in Table 1.15.2. For example, if we want to plot a data in a dashed blue line, the command becomes >> plot(y,’-b’); xlabel, ylabel The xlabel(’ter’) and ylabel(’tez?’) are used to label the x and y axes. axis The axis command sets up the limits of axes. The synopsis is aXis[2min, maz; Ymin Yaz] text The text command is used to write a text on the graphic window at a designated point. The synopsis is text(z,y,’text contents’) where 2, y locates the z,y position of the "tert contents’. In case we want to normalize the graphic window size from 0 to 1, we use text(z,y, teat contents! sc’) ginput This command allows us to pick up any point on a graphic window. The synopsis is [z,y] = ginput26 Introduction to MATLAB Chapter 1 plot(y) os} og oa 02] oat ast ast 3} plat(ty) ot oa 02} 04} as} oa} Figure 1.15.1 A Sample Plot We can pick as many points as we want on the graphic screen. The vector [z, y] then contains all the points. gtext The gtext command is used to place a text on the graphic window using the mouse input. The synopsis is gtext(‘teat’)Section 1.15 Plotting Tools 27 Table 1.15.2 Line, Mark, and Color Styles Style Line marks Color solid point. red dashed star * green dotted ° circle 0 blue b dashdot ™-.” plus + white w xmark x | invisible i Once the above command is entered or read in a function subroutine, the cursor on the graphic window is activated waiting for the mouse input, so that the “tezt’ is located at the point selected by the mouse. grid The grid command adds grids to the graphic window. It is useful when we want to clarify axis scales. An example plot constructed using some of the commands described above is presented in Fig, 1.15.2. The following commands are used for the plot output. >> t=0:0.1:20; >> plot(t, sin(t)) >> xlabel('Time(sec)’)) >> ylabel('ydata’) >> title('This is a plot example’) >> grid >> gtext('sin(t)’) >> axis([0 20 — 1.5 1.5)) pause This command is useful when we display multiple graphic windows sequen- tially, It allows us to display one at a time with the keyboard interrupt. subplot The subplot is used to put multiple plots on the same MATLAB figure window. The command is >> subplot(pgr) The plot size is adjusted by a p by q matrix on the whole size of the graphic window. Then the third index r picks one frame out of the p by q plot frames. An example subplot is presented in Fig. 1.15.3 with the following commands entered. >> 2=0:0.1:depi; >> subplot(222) >> plot(z,y) >> title(’x and y’) >> subplot(223) >> plot(z, z) = sin(2); z = cos(z);28 Introduction to MATLAB Chapter 1 This sa plot example SaaS elated, Time(s00) Figure 1.15.2 A Plot Example With Some Commands xand y 4 0) of 05 a) 5 70 xandz 1 os] | 08 a) $ 10 Figure 1.15.3 A Subplot Example >> title(’x and 2’) >> subplot(224) >> plot(z,y,'—',2,2,'-—')Section 1.15 Plotting Tools >> title(’x and ly 2}’) where pi is an internally defined variable equivalent to 7. 2CHAPTER TWO APPROXIMATION TECHNIQUES 2.1 Methods of Weighted Residual Methods of weighted residual are useful to obtain approximate solutions to a differential governing equation. In order to explain the methods, we consider the following sample problem: | aaa O<2<1 (2.1.1) u(0) =0, and u(1) =0 The first step in the methods of weighted residual is to assume a trial function which contains unknown coefficients to be determined later. For example, a trial function, fi = az(1— 2), is selected as an approximate solution to Eq. (2.1.1). Here, ~ denotes an approximate solution which is usually different from the exact solution. The trial function is chosen here such that it satisfies the boundary conditions (i.e., (0) = 0 and @i(1) = 0), and it has one unknown coefficient a to be determined. In general, accuracy of an approximated solution is dependent upon proper selection of the trial function. However, a simple form of trial function is selected for the present example to show the basic procedure of the methods of weighted residual. Once a trial function is selected, residual is computed by substituting the trial function into the differential equation. That is, the residual R becomes n= SS aye -20—aa(—2) +2 (2.1.2) Because i is different from the exact solution, the residual does not vanish for all values of z within the domain. The next step is to determine the unknown constant a such that the chosen test function best approximates the exact solution. To this end, a test (or weighting) function w is selected and the weighted average of the residual over the problem domain is set to zero. That is, 1 oo, r= f wR de = f w( G3 -a+2)de fo fo | \ da? ‘i =f w{—2a-az(1— 2) +2}d2 =0 (21.3) 5 3132 Approximation Techniques Chapter 2 Table 2.1.1 Comparison of Solution to Eq. (2.1.1) at x=0.5 a Exact Solution Collocation Least Squares Galerkin 0.0566 0.0556 0.0576 0.0568 ‘The next step is to decide the test function. The resulant approximate solution differs depending on the test function. The methods of weighted residual can be classified based on how the test function is determined. Some of the methods of weighted residual are explained below. Readers may refer to Refs [1-3] for other methods. 1, Collocation Method. The Dirac delta function, (2 — 24), is used as the test function, where the sampling point 2; must be within the domain, 0 < 2; < 1. In other words, w= 6(z— 2) (2.1.4) Let 2; = 0.5 and we substitute the test function into the weighted residual, Eq. (2.1.3), to find a = 0.2222. Then, the approximate solution becomes i = 0.2222e(1 — 2). 2. Least Squares Method. The test function is determined from the residual such that dR da Applying Eq.(2.1.5) to Eq. (2.1.2) yields w = ~2—z(1—z). Substitution of the test function into Eq, (2.1.3) results in a = 0.2305. Then i = 0.23052(1 —z). 3. Galerkin’s Method. For Galerkin’s method, the test function comes from the chosen trial function. That is, (2.1.5) di da For the present trial function, w = 2(1— 2). Applying this test function to Eq. (2.1.3) gives a = 0.2272 so that a = 0.22722(1— x). Comparison of these three approximate solutions to the exact solution at z = 0.5 is provided in Table 2.1.1. As seen in the comparison, all three methods result in reasonably accurate approximate solutions to Eq. (2.1.1). (2.1.6) In order to improve the approximate solutions, we can add more terms to the previously selected trial function. For example, another trial function is @ = a,2(1 — 2) + a9z7(1 — z). This trial function has two unknown constants to be determined. Computation of the residual using the present trial function yields R=a,(-2—24+27)+4,(2—- 62-27 +25) +2 (2.1.7) We need the same number of test functions as that of unknown constants so that the constants can be determined properly. Table 2.1.2 summarizes how to determine testSection 2.1 Methods of Weighted Residual 33 Table 2.1.2 Test Functions for Methods of Weighted Residual Method Description Collocation w; = 6(@—2)), 7=1,2,.40 where 2; is a point within the domain Least Squares w; = OR/da;, i= 1,2,...,n, where F is the residual and q; is an unknown coefficient in the trial function Galerkin w; = 04/80;, = 1,2,..,.0 where @ is the selected trial function functions for a chosen trial function which has n unknowns coefficients. Application of Table 2.1.2 to the present trial function results in the following test functions for each method. Collocation Method : wi = 6(z— 21), w2 = 6(e — 22) (2.1.8) Least Squares Method : w, = —-2—a2+27, we=2-6r—27 42° (2.1.9) Galerkin’s Method : w, =2(1—2), w2=27(1-2) (2.1.10) For the collocation method, 2; an zz must be selected such that the resultant weighted residual, ie. Eq. (2.1.3), can produce two independent equations to determine unknowns a; and az uniquely. The least squares method produces a symmetric matrix regardless of a chosen trial function. Example 2.1.1 shows symmetry of the matrix resulting from the least squares method. Galerkin’s method does not result in a symmetric matrix when it is applied to Eq. (2.1.1). However, Galerkin’s method may produce a symmetric matrix under certain conditions as explained in the next section. 4 Example 2.1.1 A differential equation is written as Lu) =f (2.1.11) where L is a linear differential operator. A trial solution is chosen such that (2.1.12) in which g; is a known function in terms of the spatial coordinate system and it is assumed to satisfy boundary conditions. Substitution of Eq. (2.1.12) into Eq.34 Approximation Techniques Chapter 2 (2.1.11) and collection of terms with the same coefficient a; yield the residual as seen below; fn R= ahs +p (2.1.18) i= Here, hj and p are functions in terms of the spatial coordinate system. Test functions for the least squares method are wy = hj, G=1,2...50 (2.1.14) ‘The weighted average of the residual over the domain yields the matrix equation : r= f wRan= Aya 4, can (2.1.15) a =1 where Ay f hihjdQ (2.1.16) In Equation (2.1.16) shows that Ajj = Aji (symmetry). { 2.2 Weak Formulation We consider the previous sample problem, Eq. (2.1.1), again. The formulation described in the preceeding section is called the strong formulation of the weighted residual method. The strong formulation requires evaluation of fj w(8?i/Ox”)dz, which includes the highest order of derivative term in the differential equation. The integral must have a non-zero finite value to yield a meaningful approximate solution to the differential equation. This means a trial function should be differentiable twice and its second derivative should not vanish. So as to reduce the requirement for a trial function in terms of order of differentiability, integration by parts is applied to the strong formulation. Then Eq. (2.1.3) becomes 1 2a =f wo (G3 -a+2)ae lo 1/ dwdi di)’ al (-Ge-wirew)ates [WE] =0 (2.2.1) As seen in Eq. (2.2.1), the trial function needs the first order differentiation instead of the second order differentiation, As a result, the requirement for the trial function is reduced for Eq. (2.2.1). This formulation is called the weak formulation. Weak formulation has an advantage for Galerkin’s method where test functions are obtained directly from the selected trial function. If a governing differential equa- tion is the self-adjoint operator, Galerkin’s method along with the weak formulationSection 2.3 Piecewise Continuous Trial Function 35 Tet Ties i, Figure 2.3.1 Piecewise Linear Functions Figure 2.3.2 Piecewise Linear Trial Function results in a symmetric matrix in terms of unknown coefficients of the trial function. Using a trial function # = az(1—) for the weak formulation, Eq. (2.2.1) results in the same solution as obtained from the strong formulation as expected. However, when a piecewise function is selected as a trial function, we see the advantage of the weak formulation over the strong formulation. 2.3 Piecewise Continuous Trial Function Regardless of the weak or strong formulation, the accuracy of an approximate solution so much depends on the chosen trial function. However, assuming a proper trial function for the unknown exact solution is not an easy task. This is especially true when the unknown exact solution is expected to have a large variation over the problem domain, the domain has a complex shape in two-dimensional or three- dimensional problems, and/or the problem has complicated boundary conditions. In order to overcome these problems, a trial function can be described using piecewise continuous functions. Consider piecewise linear functions in an one-dimensional domain as defined below: (e-aia/he forma Sasa i(2) = 4 (tig —2)/higr for ay So < aig: (2.3.1) 0 otherwise The function defined in Eq. (2.3.1) is plotted in Fig. 2.3.1 and Example 2.3.1 illustrates the use of the function as a trial function.36 Approximation Techniques Chapter 2 & Example 2.3.1 Consider the same problem as given in Eq. (2.1.1). It is rewritten here @u #4 -us-2, 0<2<1 4 { (0) = 0, and u(1) = 0 (2.1.4) The weak formulation is also rewritten as below: oe Pi. t= o( G3 -a+2)ée =[ ee az +[v%]' =o (2.2.1 =/, de dg 7 Witew “al,= 1) A trial function is chosen such that d = a1¢1(z) + a2@0(z) in which ay and a2 are unknown constants to be determined, and ¢; and ¢o are defined as below: 3a, or(z) = { 2-32, (2.3.2) canta © IAIAIA Rae IAIAIA Heontaim 0, $2(z) = { 32-1, (2.3.3) 3-32, cansame © IAINIA eas IAIAIA onto gi(x) and ¢o(z) are plotted in Fig. 2.3.2. For the present trial function, the problem domain is divided into three subdomains and two piecewise linear functions are used. Of course, more piecewise functions can be used along with more subdomains to improve accuracy of the approximate solution. The trial function can be rewritten as a,(3z), O= (4.2.1) Rewriting Eq. (4.2.1) gives (4.2.2) As a result, the equivalent spring constant for the axial bar is bey = (4.2.3)Section 4.2 Axial Member 15 Uy Us Figure 4.2.1 Axial Members Represented by Equivalent Springs & Example 4.2.1 Axial members can be represented by serial and/or parallel linear springs. For example, a bar of telescoped shape can be replaced by a series of springs as shown in Fig. 4.2.1. There are three linear spring elements and each element has matrix expression as shown in Eq. (4.1.4) with proper spring constant kj. Assembling them gives ko ky 0 0 uy fi -k kth -b 0} )ul_)o 0 ky ketks —ks| ) us f=) 0 (42.4) 0 0 —kg ks U4 P 4;2: and fy is the unknown reaction force at the left end support. Instead, u; = 0 is given as the boundary condition. Solution of Eq. (4.2.4) with this boundary condition will result in f = —P which can be also obtained from the static equilibrium. However, the finite element formulation already includes equations of static equilibrium so that we may not use additional equilibrium equations. These equations are redundant. & Example 4.2.2 Consider a statically indeterminant system as shown in Fig. 4.2.2, The axial member can be replaced by two linear springs as seen in the figure. The finite element matrix equation for this system is 0.5k —0.5k 0 wy fi 05k 15k -k]iu}= 20 (4.2.5) 0 -k kJ lus fs where k = 42 and f; and f3 are the unknown forces at the supports. Because the system is statically indeterminate, we cannot find the forces from the static76 Direct Approach With Spring System Chapter 4 Dy) Migr peat 4 Uy Ue us 2L L Figure 4.2.2 A Statically Indeterminate Axial Member y AE yk “AE Y P be | P a ee / Rigid Plate Figure 4.2.3 A Spring Representing a Statically Indeterminate System equilibrium equation only. However, the finite element formulation includes not only equilibrium but also geometric compatibility (compatibility of deformation). Therefore, Eq. (4.2.5) along with boundary conditions u; = u3 = 0 can solve the deformation as well as the reaction forces. Another statically indeterminate system made of axial rods can be represented by a spring constant as shown in Fig. 4.2.3. tSection 4.4 Other Systems 17 Figure 4.3.1 Torsional Members 4.3 Torsional Member A circular rod subjected to a twisting moment produces an angle of twist TL "a (43.1) in which @ is the angle of twist, T is the applied torque, L is the length of the member, G is the shear modulus, and J is the polar moment of inertia of the circular cross- section. Rewriting the equation yields GI T=6 (4.3.2) ‘The equivalent spring constant is keg = $2. Torque T corresponds to the spring force F and angle of twist @ to the spring displacement u. As examples, both statically determinate and indeterminate torsional members are shown in Fig. 4.3.1 along with their equivalent spring systems. 4.4 Other Systems Other frequently used engineering systems, which can be easily substituted with a spring system, are heat conduction, simple fluid flow along pipes and electric cir- cuits. For one-dimensional heat conduction, heat flux is proportional to temperature78 Direct Approach With Spring System Chapter 4 ky ky —— i — — ky ih Figure 4.4.1 Mechanical Forces From Newton’s Third Law difference. That is, heat flux q is q=—-kAT (4.4.1) where k; is the heat conduction coefficient and AT is the temperature difference between two end points of a one-dimensional bar. The minus sign in the equation denotes heat flux is in the opposite direction as the temperature increases along the positive axis. The equivalent spring system has the spring constant key = —ke, the spring force F = q and the spring displacement u = T. Fluid flow rate through a pipe of constant cross-section is proportional to the pressure difference of the two ends. That is, Q=~-kpAp (442) Here, Q is the flow rate, Ap is the pressure difference, and kp is the proportional constant. For a laminar flow, the proportional constant can be expressed as __ndt ~ 128nL where yp is the fluid viscocity, d is the pipe diameter, and L is the pipe length. The equivalent spring system has the spring constant keq = —kp, the spring force F = Q and the spring displacement u = p. Electric current flow é through a resistance R is ae i=R (44.4) where V is voltage. The equivalent spring system has spring constant key = 4, force F = i, and displacement u = V. One may think what about V =i R such that F =V, key = Rand u=i. Which one is correct? Is the equivalent spring constant Ror #? One is right and the other is wrong. One way to select the right form is to understand the nature of spring force and to find the parameter equivalent to the force. Then, the rest of parameters can be determined accordingly. When two springs are separated each other, internal forces occur between the two springs. These forces are equal in magnitude and opposite in direction. When the two springs are put together, the forces cancel each other and become zero if there is no external force applied at the joint as seen in Fig. 4.4.1. This is known as Newton’s third law. When considering electric current, its sum at the joint of resistants (i.e. the middle point in Fig. 4.4.2) is also zero like the force. However, voltage does not vanish at such a joint. As a result, electric current is similar to the mechanical force. ‘Therefore, the equivalent spring constant for the electric circuit is 4 but not R. kp (4.4.3)Section 4.4 Other Systems 79 k = 16 Figure 4.4.2, Flow Through a Pipe System # Example 4.4.1 Consider a pipe system for fluid flow as seen in Fig. 44.2, The pressure at the inlet is 200 while the flow rate at the outlet is 50. The proportional constants kp in Eq. (4.4.2) are given in the figure. All units are assumed consistent. We want to determine the flow rate between node 3 and node 4. Using the linear spring equivalency, the matrix equation becomes -1 1 0 0 0 O4fm a i is ¢ 3 6 0 lle 0 0 4 -2 2 16 0 | Jos 0 oer UCU 0 (445) 0 0 1 1 49 32 | |p 0 0 0 0 o 32 —32d bog Qo = 50 Applying the known pressure pi = 200 to this equation yields the following pressure at each node. P2 = 150, ps = 142.2, pa = 147.6, ps = 139.6, po = 138 ‘Therefore, the flow rate between nodes 3 and 4 is Qs-4 = —2 x (147.6 — 142.2) = -10.8 Hence, the flow is upward with rate of 10.8. t80 Direct Approach With Spring System Chapter 4 Figure P4.1 Problem 4.1 41 42 43 44 45 46 Problems Find the system of equations for the spring system shown in Fig. P4.1. Solve the matrix equation to find the displacements of the nodal points. A circular shaft is made of two different materials and fixed at both ends. It is subjected to torsions as shown in Fig. P4.2. The diameter of the shaft is 0.1 m. Find the angles of twist at the nodal points. For the given mass-spring system (see Fig. P4.3), (a) develop the system mass and stiffness matrices to determine natural frequencies of the system. (b) Apply the given boundary conditions to the matrix equations. An electric circuit is shown in Fig. P44, Find the current flow using the equivalent spring system. For a laminar pipe flow, the flow rate is proportional to pressure difference. Construct the system of equations for the given pipe flow as shown in Fig. P4.5 in order to find the flow rate through each pipe. Heat conduction through a circular pipe is expressed as ___2kL 1 =~ inro/ny in which & is the heat conduction coefficient, L is the length of the cylinder, and 7) and r; are the outer and inner radii of the cylinder, respectively. Therefore, two concentric cylinders with radii r, < rg < rg can be represented by two springs in serial connection. Find the equivalent spring constants for the two linear springs in Fig. P4.6 and construct the system of equations.Problems 81 22 KN-m 22 KN-m — — Fixed elem #1 elem #2 | elem #3 | Fixed 2m en 1.5m Diameter = O01 m Element 1 & 2: G = 80 GPa Element 3: G = 40 GPa Figure P4.2 Problem 4.2 rigid bar igi ks k a: i 4 downs } WinWi-Tap Lt 6 fin —t rigid bar 1 [ue Lug Leu [> Us Figure P4.3 Problem 4.3 -—> u. 10 volts Figure P4.4 Problem 4.4Chapter 4 Direct Approach With Spring System 82 Figure P4.5 Problem 4.5 BES ERROR % Re BS see SO x < > SS =] y2-os vs— un now up (5.2.4) or a3 23-22 2-23 22-21 us where 1 lan A= 5det|1 2 wn (5.2.5) 1 ts Ys Magnitude of A is equal to the area of the linear triangular element. However, its value is positive if the element node numbering is in the counter-clockwise direction andSection 5.2 Linear Triangular Element 87 (=3, v3) Uy (=, %) (22, Y2) Figure 5.2.1 Linear Triangular Element negative otherwise. For the finite element computation, the element nodal sequence must be in the same direction for every element in the domain. Substitution of Eq. (8.2.4) into Eq. (5.2.2) produces = Ay(z,y)uy + Ho(2, y)ue + a(x, y)us (5.2.6) in which H(z, y) is the shape function for linear triangular element and it is given below: H,= Filles — zay2) + (y2 — ya)z + (29 — 22)y] (5.2.7) Hy = gilleam — 198) + vs wd + (#1 —25)al 628) Hs= Fille — ays) + (yr — yo) 2 + (22 — 21)y] (5.2.9) These shape functions also satisfy the conditions Hi(xj,93) = 85 (5.2.10) and 3 Vaal (5.2.11) ia Here, 6;; is the Kronecker delta. That is, {a i ay (6.2.12) A problem domain is discretized into a number of finite elements using the linear triangular elements. An example of finite element mesh discretization is illustrated in Fig. 5.2.2. As seen in the discretization, the actual curved boundary is approximated by a piecewise linear boundary. The crude mesh in the figure may be refined for88 Laplace’s and Poisson’s Equations Chapter 5 Figure 5.2.2 Finite Element Discretization closer approximation of the actual boundary using linear triangular elements. Another alternative is to use higher order finite elements which can fit the curved boundary using higher order of polynomial expressions. For a linear triangular element shown in Fig, 5.2.1, the element matrix is computed as derived below. oh a f (8m du, Owdu a wie f (Geet Seseane J [ie a \¢ os atte 4p y)an (5.2.13) E a € + where 9° is the clement domain. Performing integration after substituting the shape functions Eqs (5.2.7) through (5.2.9) into Eq. (5.2.13) gives ki kia kis [k= [ kao ‘| (5.2.14) ka: a2 koa in which ku = Filles — 22)? + (y2 — ys)" (5.2.15) k= Filles ~ ©2)(21 — 23) + (ye — ys)(ys — v1)] (5.2.16) his = Ules — 22)(22— 21) + (y2 — ¥s)(¥1 — ¥2)] (6.2.17) kay = hie (5.2.18)Section 5.2 Linear Triangular Element 89 (0,1) (0,0) (1,0) Figure 5.2.3 Triangular Domain With One Element 1 kan = Gyl(ti — 23)” + (vs — n)*)] (5.2.19) 1 bas = Galler ~ #5)(22 — 21) + (vs — 1)(v1 — ¥2)] (5.2.20) kay = his (5.2.21) kso = kos (5.2.22) 1 : . ss = gyl(e2 — 21)? + (1 ~ v2))) (5.2.23) Because 42 and 9! are constant for the linear triangular element, the integrand in Eq. (5.2.18) is constant. As a result, the integration in Eq. (5.2.13) becomes the integrand multiplied by the area of the element domain and the result is given in Eqs (5.2.15) through (5.2.23). & Example 5.2.1 We compute the element matrix for Poisson’s equation for the linear triangular element shown in Fig. 5.2.3. Element node numbering is in the counter-clockwise direction. The area of the triangular element is 0.5 and the element matrix is 10 -0.5 -05 [k]=]-05 05 0 (5.2.24) 05 0 05 ‘The other domain integral term to be evaluated in Eq. (5.1.18) is f wa(z,y)Q (5.2.25) In90 Laplace’s and Poisson’s Equations Chapter 5 This integration results in a column vector as shown below. Computation of this integral over each linear triangular element yields Ay if {ie} xen (5.2.26) ine | Hy Analytical integration may not be simple depending on function g(z,y). Then, a numerical integration technique may be applied to compute this integral. Some numerical techniques are discussed in Chapter 6. 5.3 Bilinear Rectangular Element The bilinear rectangular element is shown in Fig. 5.3.1. The shape functions for this element can be derived from the following interpolation function: u= a; +92 + agy + ay (5.3.1) This function is linear in both z and y. Applying the same procedure as used in the previous section results in m= Eb-2He-v) 6.8.2) ie xe Tae 9) (5.3.3) Age Hor avlery) (5.3.4) Ha= xe ~2)(e+y) (5.3.8) where 2b and 2c are length and height of the element, respectively. The shape functions Eqs (5.3.2) through (5.3.5) can be obtained by product of two sets of one-dimensional shape functions. Let the linear shape functions in the «-direction with nodes located at = —b and z = b be dle) = 20-2) (6.3.6) and d2(z) xt A) (5.3.7) Similarly, the linear shape functions in the y-direction are he) = Ze-9) 638) and vale) = E(e+) (5.3.9)Section 5.3 Bilinear Rectangular Element 91 Figure 5.3.1 Bilinear Element Product of Eqs (5.3.6) and (5.3.7) and Eqs (5.3.8) and (5.3.9) yields Eqs (5.3.2) through (5.3.5). The shape functions obtained by products as shown above are called the Lagrange shape functions. @ Example 5.3.1 We want to compute the element matrix for Poisson's equation using the bilinear shape functions. oy ft de re) — Or 4 5H: 1H s ula f () ae tee Ye ae id 2: oat, Ox oH, by Ha y oH, 8, @Hx 98s) \ a0 5.3.10) ans ( U'ey Oy ot) (5.3.10) ai ey where H; is the bilinear shape function. This will be a matrix of 4x4, The first component is OH, OH, + oh OW, wie [) Lae ee + a ae ane = LL agizalle— 0? + vaya _ e+e ~ 8be Performing integrations for all terms results in the following element matrix for the bilinear rectangular element. (5.3.11) ir kia hig hia ey. | Az boo kos hog [kK] = be is is is (5.3.12) hig koa kaa haa92 Laplace’s and Poisson’s Equations in which BP+e? k= abe BF — 2c? re bbe e+e? hig = She co? — 26? hua = kag = hy kos = kia koa = his, egg = yy kes = kia kag = yy t The other domain integral becomes b A ola, ja H, [Up vidude Hy This is similar to that in Eq. (5.2.26). 5.4 Boundary Integral The boundary integral in Eq. (5.1.18) is au au [egar=D [wie Chapter 5 (6.3.13) (5.3.14) (5.3.15) (5.3.16) (5.3.17) (5.3.18) (5.3.19) (5.3.20) (5.3.21) (5.3.22) (5.3.23) (5.4.1) where subscript n denotes the natural boundary and superscript ¢ indicates the element boundary. Here, the summation is taken over the elements which are locatedSection 5.4 Boundary Integral 93 BK =gwen Qu =piven Figure 5.4.1 Elements at Boundary at the boundary of the domain and whose element boundaries are subjected to the natural boundary condition as shown in Fig. 5.4.1. For simplicity, we consider an element boundary which is parallel to the z-axis as seen in Fig. 5.4.2. The element boundary is subjected to a positive constant flux. That is, the flux is in the outward direction which is assumed to be positive. Since linear triangular elements are used for the domain discretization, the element boundary has two nodes as shown in Fig. 5.4.2. As a result, linear one-dimensional shape functions are used to interpolate the element boundary. The boundary integral along the element boundary becomes (5.4.2) where hig = 2} — 2; :length of the element boundary (5.4.3) This colume vector is added to locations associated with nodes i and j. If the element boundary is along the y-axis or is in an arbitrary orientation about the zy-axes, the result is obtained as long as i; is the length of the element boundary. & Example 5.4.1 We consider a heat conduction problem with a triangular shape of domain which is discretized into four linear triangular elements (see Fig. 5.4.3). There are six nodes in the domain. One boundary is insulated or symmetric so that there is no flux (34 = 0) through the boundary. Another boundary has a constant heat flux and the third boundary has a known temperature. Find the temperature at the nodal points. Each element matrix can be obtained from Eqs (5.2.14) through (5.2.23). The local and global node numbering is shown for each element in Fig. 5.4.4. The94 Laplace’s and Poisson’s Equations Chapter 5 ‘J=constant Figure 5.4.2 Triangular Element Subjected to Constant Flux global node numbering is used to identify what nodes are associated with each element while the local node numbering is related to the assignment of shape functions to the nodes. Therefore, the local node numbers are always 1, 2 and 3 for the linear triangular clement. For the present elements, element matrices are the same and one of them is given below: 05 -0.5 0.0 {K]}=|-05 10 -05 (5.4.4) 0.0 -05 05 If the local node numbering changes for each element, the element matrix becomes different from that in Eq. (5.4.4). Assembling the element matrices into the system matrix based on the global node numbers results in 05 -05 00 00 00 00 0.5 20 -10 -05 00 00 00 -10 20 00 -10 00 00 -05 00 10 -05 0.0 0.0 00 -1.0 -05 20 -0.5 00 00 00 00 -05 05 [K]= (5.4.5) for the system nodal vector { u; Uz Us U4 Us ts }. ‘The system column vector is obtained from the boundary integration of the given flux. Element boundaries with specified nonzero flux are boundaries 4 — 5 and 5 —6. Using Eq. (5.4.2), the equivalent nodal fluxes are (8-0) eso (8)-(3 een andSection 5.4 Boundary Integral 95 Figure 5.4.3 A Triangular Domain On the other hand, element boundaries 1 ~ 3 and 3 — 6 have zero nodal fluxes because they are insulated. Combining all these vectors yields the system column vector {F}={F Fh 0.0 Fy 4.0 2.0}7 (6.4.8) where Fy, Fp and F4 are unknown nodal fluxes. At node 4 flux is known at the edge 4 — 5 but it is not known at the edge 2— 4. The nodal flux at node 4 is affected by fluxes at the both edges and one of them is not known. As a result, Fy is unknown. The same explanation holds for F,. Since temperature is known at nodes 1, 2 and 4, we can solve the matrix equation [K]{u} = {F} (5.4.9) with tu = 0.0, uz = 0.0 and ug = 0.0. The solution gives us = 3.0, us = 6.0 and ug = 10.0. ¢ 4 Example 5.4.2 One common boundary condition in heat transfer is heat convection at the boundary. This boundary condition is expressed as Ou rae a(u— to) (5.4.10) where @ is the heat convection coefficient and u, is the ambient temperature. ‘That is, heat flux is proportional to the temperature difference of the body surface and the environment. Rewriting this in a more general expression gives & =autb (5.4.11)96 Laplace’s and Poisson’s Equations Chapter 5 global node # / local node # AAVA element ° Cen 7 element #3 aan 7 Figure 5.4.4 Elements With Local and Global Node Numbers where @ and b are known functions because Us is a known value, Substituting Eq. (5.4.11) into the element boundary integral in Eq. (5.4.1) results in f . wotar = iE wia(z, yu + (2, y)}dP (5.4.12) Whenever there is a product of test function and trial function, the term becomes a matrix while the test function only produces a vector. As a result, the first term of Eq. (5.4.12) becomes using linear shape functions to interpolate the element boundary yar { a (6.4.13) uy where s; and s; are the coordinate values of the local axis located along the element boundary as seen in Fig. 5.4.4, and u; and u, are the nodal variables at the element boundary. This integral results in a matrix of 2 x 2 for an element with two nodes on the boundary. This matrix should be added to the system matrix. The rest term in Eq. (5.4.12) can be dealt in the same way described in this section { 5.5 Transient Analysis The governing equation for transient heat conduction is 24 9? & L(+ *) inQ (5.5.1) where ¢ denotes time and a is a known function. Usually, for heat conduction problems with constant material properties a is equal to “€ where k is the coefficient of heatSection 5.5 Transient Analysis 97 conduction, p is density and cp is specific heat. Here, heat. generation or heat sink is neglected. Applying the method of weighted residual to Eq. (5.5.1) in the same way as given in Section 5.1 gives 84 a9 dw du , dw du 1 6u t= [we +2 (Fests Bat og au) aa a 5 ar (6.5.2) As noticed here, the method of weighted residual is applied to the spatial domain but not to the temporal domain regardless it is a steady state or a transient problem. As a result, the difference between the transient and steady state problems is the first term in Eq. (5.5.2). The other difference is, of course, the variable u is a function of both space and time for the transient problem. The variable u = u(z,y,t) is interpolated within a finite element in a similar way as before using shape functions. ulesnt) = Mile, wus) (65.3) : where H;(z, y) is the shape function and n is the number of nodes per element. One thing to be noted here is that the shape functions are used to interpolate the spatial variation within the element while the temporal variation is related with the nodal variables. Applying Eq. (5.5.3) to the first term in Eq. (5.5.2) yields A i [ue] = i He \{Hi Hy He}dQd ty (55.4) i Las re for a linear triangular element. On the other hand, the matrix and vector obtained from the second and third integrals of Eq. (5.5.2) are the same as those developed in the previous sections other than that a should be included in the matrix. Computation of Eq. (5.5.4) results in eteeeneceas mi=4 : 2 | (5.5.5) eee iecan where A is the area of the triangular element. Similarly, the bilinear rectangular element yields 4212 2 eee 36/12 4 2 65.6) 2124 ‘Therefore, the final matrix equation for Eq. (5.5.1) becomes (M]{a}' + [K]{u}! = (FY (5.5.7)98 Laplace’s and Poisson’s Equations Chapter 5 Because this equation should be true at any time, we put superscript ¢ in Eq. (5.5.7) to denote the time when the equation is satisfied. Furthermore, matrices [M] and [K] are independent of time. Now, the parabolic differential equation has transformed into a set of ordinary differential equations using the finite element method. In order to solve the equations, we use the finite difference method for the time derivative. The next sections show the solution techniques. 5.6 Time Integration Technique First of all, we explain the forward difference method for time derivative. The forward difference is expressed as oye {upttat —{u}? wee (5.6.1) Substitution of Eq. (5.6.1) into Eq. (5.5.7) results in [M]{u}*4* = Ae({r} — fA] {u}) + [Ml {uy* (5.6.2) In the above equation, all the terms defined at time ¢ are put on the right-hand side of the equation while the term associated with time ¢ + At is at the left-hand side, Equation (5.6.2) can be solved from the given initial condition {u}° and known boundary conditions {F}* as explained below: 1. Setting t = 0 in Eq. (5.6.2) can find the solution for {u}4* from {u}° and {F}°. 2. Once {u}4* is found, we can continue the previous step again by setting t = A¢ in Eq. (5.6.2) in order to determine {u}?4'. This step is repeated until the solution reaches the final time. The forward difference technique in Eq. (5.6.1) has the local truncation error O(At?) and the global truncation error O(At) where O denotes the order of error. The forward difference technique is conditionally stable so that a proper size of time step At should be used to have a stable solution. ‘The next technique is the backward difference technique. For this technique, Eq. (5.5.7) can be rewritten at time t + At. [MI{u}'t4t + [K]{u}t4! = (Fy +4t (5.6.3) The time derivative in the backward difference is payee = oO Go an Use of Eq. (5.6.4) with Eq. (5.6.3) results in ([M] + Ae(K]){u}*4t = Aetry +4! + [M}{u}* (6.6.5)Section 5.6 Time Integration Technique 99 The solution procedure is similar to that for the forward difference technique. The local and global truncation errors are also the same as those for the forward difference technique. However, the backward difference technique is unconditionally stable. Therefore, any size of At can be used without worrying about stability. However, the time step size is, of course, important for accuracy because of the truncation error. The other technque is the Crank-Nicolson method. For this technique, we write Eq. (5.5.7) at time t + 4¢ instead of t. Then, [Muy + [uy ® = (rye (5.6.6) The time derivative term is expressed using the central difference technique like jt {u}ttd* — {apt xi (5.6.7) On the other hand, the other terms are computed as average like tate = Supt + tuys) (668) and (ryt = Lccey + (ry) (63) Substitution of Eqs (5.6.7) through (5.6.9) into Eq. (5.6.6) yields (2(M] + At[K]){u}'t4t = An({F} + (FF +4) + (2[M] - At{K]{u}! (6.6.10) ‘The Crank-Nicolson method is also unconditionally stable and the global truncation error is O(At?) so that it is one-order higher than the other two techniques. 4 Example 5.6.1 Let us solve the following set of ordinary differential equations using the backward difference method. [M]{u} + [K]{u} = {F} (5.6.11) where 2100 ot M=slo iad (5.6.12) ot ed) -1 2 -1 0 ele (5.6.13) 0 0 -1 1100 Laplace’s and Poisson’s Equations Chapter 5 and {F}={Fi 00 Fi}? (5.6.14) Here, F, and Fy are unknown while u; = 100 and tq = 100 are known as boundary conditions. In addition, the initial condition states {u}° = 0. Substitution of Eqs (5.6.12) throngh (5.6.14) into Eq. (5.6.5) yields d4At 2-At 0 0 ee tat Pyoat t-ar 0 uptat 0 Gm At 4426 i oe (ee 0 0 fat §4+Ael Cujtet dui + qua+ Fit dul + 2uh + #4 guy + 5g + uy duh + Bul + Fat Applying the boundary conditions with At = 1 as the time step size, Eq. (5.6.15) (5.6.15) becomes ty ute 100 -§ 8 _§ 9 ubtat Lyt 4 2uf 4 Lut 5 os 2 a=) gut gua + ous oF wel yagtar =) fab + dug + bad oe 0 0 0 1 uatat 100 Let t = 0 in Eq. (5.6.16) and use the initial condition to find the solution at t= 1. The solution is {ul ub ub wl}={100 45.5 45.5 100} (5.6.17) where superscript 1 denotes the solution at time ¢ = 1. To continue the solution, let t = 1 in Eq. (5.6.16) and use the previous solution in Eq. (5.6.17). Then, the solution at time t = 2 is {u? uZ u2 w2}={100 75.4 75.4 100} (5.6.18) ‘This process continues until the final time. As expected, the solution approaches to the steady state of uniform value of 100. { 4 Example 5.6.2 Let us solve Example 5.6.1 using the Crank-Nicolson method. Applying Eqs (5.6.12) through (5.6.14) to Eq. (5.6.10) yields 24at }-At 0 0 Chey 3-At $+2At 4-At 0 ubtat | 0 oA fas gaat] Y agtar 0 = 0 0 g-At 3+Atl Luft 0 0 us t4+At 0 al ano jtat us 44At $-Atl Caf At(F} + Fit4*) 0 : (6.6.19) Ai (Fi + Fi*4*)Section 5.7 Axisymmetric Analysis 101 Figure 5.7.1 Cylindrical Coordinate Applying the boundary conditions uj = 100 and u4 = 100 with At = 1 results in the following matrix equation. 1 0 0 9 tee 100 7 0 Hat age ae = oe us guj — 2uh + fu! ye z. = i ? ; 5.6.20 Wp 3) agra =) ab da dh i O00 1 utr 100 Applying the initial condition with ¢ = 0 yields the solution at time t = 1. {ul ul uf ul}={100 25 25 100} (5.6.21) ‘The solution at the next step, ie. t= 2, becomes {uz u} ug ul} ={100 81.3 81.3 100} (5.6.22) using the solution in Eq. (5.6.21) and t = 2 for Eq. (5.6.20). ¢ 5.7 Axisymmetric Analysis Laplace’s equation in the cylindrical coordinate system is written as below: au pa 9 (57.1) oe Or? where r, ¢ and z are the radial, circumferential, and axial axes, respectively, as shown in Fig. 5.7.1. For the axisymmetric problem, variable w is independent of the circumferential axis ¢. This is the case where the domain is axisymmetric and all the described loading and/or boundary conditions are also axisymmetric. Therefore, the governing equation is simplified to Ou lou, eu or * sort O27 — 9 (67.2) for the axisymmetric analysis.102 Laplace’s and Poisson’s Equations Chapter 5 Let us apply the weighted residual method. The integral becomes au, 1du, Oe fo $1 4S aaa (8.7.3) The first two terms in Eq. (5.7.3) can be rewritten as 10 a Lota 3) toa Now, the domain integral can be expressed as [sosan= ff [ toarasiras =2n i f rf(r2)drdz (5.7.5) }an (6.7.4) where f(r, 2) is any function which is independent of ¢. Applying Eq. (5.7.5) to Eq. (5.7.4) gives arf fw {Z(-$) +-Sz}an (5.7.6) ‘The weak formulation of Eq. (5.6.7) using the integration by parts becomes dw du, dw du du ~on f f+ (3 oe + eee aed + fru Star (6.7.2) where the boundary integral is on the rz-plane and n is also the outward normal unit vector to the boundary. Equation (5.7.7) is now expressed in terms of the radial and axial axes, ie. r and z. As a result, we need a finite element discretization in the rz-plane like a two-dimensional analysis. The same kinds of shape functions can be used for both two-dimensional and axisymmetric analyses of Laplace’s equation. However, there is one difference between the two formulations. The axisymmetric analysis contains r within the integral while the two-dimensional analysis does not include r. Let us use the linear triangular element for the axisymmetric analysis. The element matrix for the triangular element can be written as wma f (( ce + Sia) (oh Oe fp }) are (6.7.8) a $BSection 5.8 Three-Dimensional Analysis 103 (tq) 23) ty 22) (ty 24) r Figure 5.7.2. Triangular Axisymmetric Element where H; is given in Eqs (5.2.7) through (5.2.9) with replacing x and y by r and z. The element is shown in Fig. 5.7.2. As discussed in Sec. 5.2, #4 and 44 are independent of r and z. We also know that [ / rdrdz = Are (5.7.9) fe where A is the area of the triangular element as defined in Eq. (5.2.5), and re = 3(ry +12 +1) is the r coordinate value of the centroid of the triangle as seen in Fig. 5.7.2. Consequently, the element matrix for the linear triangular axisymmetric element is ki kia kis [K*] =2nr, | kor koe kes (5.7.10) ks kaa kag in which kj; is the same as given in Eqs (5.2.15) through (5.2.23) except that 2; and ui are replaced by r; and 2 for the axisymmetric analysis. ‘The flux at the boundary is also handled in the similar way as the two- dimensional analysis. However, the boundary integral for the axisymmetric analysis also contains r. If there is a uniform flux on the boundary of a linear triangular element for the axisymmetric problem as shown in Fig. 5.7.3, the equivalent nodal flux vector becomes rFgl{1 1}7 where # = 3(r;+1j) is the average r coordinate value of the two boundary nodes é and j, q is the value of uniform flux per unit area, and / is the side length of the element as seen in Fig. 5.7.3. 5.8 Three-Dimensional Analysis For the three-dimensional analysis of Poisson’s equation, Eq. (5.1.18) can be extended directly into du Ou, Owdu , dw du du ram f Fer + ere + Fee an f walen.2)an+ [wear (6.8.1)104 Laplace’s and Poisson’s Equations Chapter 5 wral Figure 5.7.3. Flux on Axisymmetric Element After discretization of the domain into finite elements, we can compute element matrices and vectors as before. For further explanation, we use a tetrahedral element as shown in Fig. 5.8.1. The variable interpolation for this element is assumed u={X}7{c} (5.8.2) where {C}={er co 03 c}" (5.8.3) and {xX}={1 2 y z}? (5.8.4) That is, the interpolation function is assumed to be linear in terms of every axis. Evaluation of the variable at the nodal points gives uw lmym aye el ele (68.5) ug 1 zg ys 23] ) ¢3 U4 1 ozq yg zd Leg or Eq. (5.8.5) can be put in the following way {u} = [X]{C} (5.8.6) in which : {u}={u uz us wu} (5.8.7) After taking the inverse of matrix [X] and pre-multiplying it to both sides of Eq. (5.8.6), we substitute the resulting expression into Eq. (5.8.2). Then we obtain w= {X)"RP a) = 17 (a) (68.8) where {HJ ={Hi Ho Hz Ha} ={X}"[X]? (5.8.9)Section 5.8 Three-Dimensional Analysis y (qty %) eg (%y Vy %) 7s oe @ J ou WY, Fy Mes %) z Figure 5.8.1 Tetrahedral Element are the shape functions for the tetrahedral element with four nodes. 105 Substitution of the shape functions into Eq. (5.8.1) with element discretization results in an element matrix which is oH, 2 OH, tt bz OH; OH; wel= f () ae pe ne ac oHs € s 7 afseBelget 411 12 413° 14 [xy} = [a2 022 25 ane 431 432 433 G34 44) 442 443 O44 Then, the shape functions can be expressed as Ay(2,y,2) = a1 + ane + aay + an2 Ha(z,y, 2) = ay2 + az2z + agoy + Ganz Hg(x,y, z) = ay3 + ao32 + aggy + @432 Ha(z,y, 2) = a14 + azqz + agay + a44z Inserting Eqs (5.8.12) through (5.8.15) into Eq. (5.8.10) yields ku hia his kag kor koa kas koa kay ks2 kag kaa kay kan ag hag [K‘]}=V Co os OHay oy (5.8.10) (5.8.11) (5.8.12) (5.8.13) (5.8.14) (5.8.15) (5.8.16)106 in which Laplace’s and Poisson’s Equations kur = (an)? + (aan)? + (aan)? hz = G42 + 43192 + 41042 hig = a21a23 + @31033 + 441043 Big = 21424 + 431094 + 041045 kay = kia kop = (a22)? + (ase)? + (az)? kos = az2d23 + 432033 + G42043 kag = Qo2a24 + 432034 + a42044 kay = is, kegp = kas kg = (a25)? + (as)? + (ass)? aq = @zadz4 + G33034 + 43044 ka = kis a2 = koa kag = kaa kag = (az4)? + (34)? + (asa)? Furthermore, V is the element volume. ‘The flux boundary condition on the three-dimensional analysis can be treated in the following way. For a uniform flux on one side of the tetrahedral element, the flux column vector becomes 4¢2{1 1 1}7 for the nodes on the element boundary as shown in Fig. 5.8.2. Here, A, is the surface area of the element side on which a uniform flux of q is applied. The element matrix for the transient term (i.e. $4) is w= 5 heee Hewes 1 1 1 2. Beee Chapter (5.8.17) (6.8.18) (5.8.19) (5.8.20) (6.8.21) (5.8.22) (5.8.23) (5.8.24) (5.8.25) (5.8.26) (5.8.27) (5.8.28) (6.8.29) (5.8.30) (5.8.31) (5.8.32) (5.8.33)Section 5.9 MATLAB Application to 2-D Steady State 107 Figure 5.8.2. Triangular Boundary With Constant Flux u=100sin( TE Figure 5.9.1 Mesh With Linear Triangular Elements for the tetrahedral element. 5.9 MATLAB Application to 2-D Steady State Analysis This section shows some examples for two-dimensional steady state problems using MATLAB programs. Both linear triangular and bilinear rectangular elements are used.108 Laplace’s and Poisson’s Equations Chapter 5 & Example 5.9.1 We want to solve the two-dimensional Laplace equation for the following given conditions. The domain and the finite element discretiza- tion is shown in Fig. 5.9.1. au | tu bz? * By? for 0 < 2 <5 and 0 < y < 10. The boundary conditions are u(z,0) = 0 for 0 <2 < 5, u(0,y) = 0 for 0 < y < 10, u(2,10) = 100sin(x2/10) for 0 <2 <5, and “5M = 0 for 0 < y < 10. The MATLAB main program along with function programs are listed below. Some function programs listed in previous chapters are not listed here. Appendix A liste all the function files. %- % EX5.9.1.m % to solve the two-dimensional Laplace equation given as % u,xx + wyy =0,0 node no. and j-> x or y — gcoord(1,1)=0.0; gcoord(1,2)=0.0; 0 (6.9.1)Section 5.9 MATLAB Application to 2-D Steady State gcoord(2,1)=1.25; geoord(2,2)=0.0; .75; gcoord(4,2)=0.0; gcoord(5,1)=5.0; gcoord(5,2)=0.0; gcoord(6,1 coord (6,2 gcoord(7,1)=1.25; gcoord(7,2)=2.55 gcoord(8,1)=2.5; gcoord (8,2 gcoord(9,1)=3.75; gcoord(9,2)=2.5; gcoord(10,1)=5.0; gcoord(10,2 gcoord(11,1)=0.0; gcoord(11,2 gcoord(12,1)=1.25; gcoord(12,2)=5.0; gcoord(13,1}=2.5; gcoord(13,2) gcoord(14,1)=3.’ gcoord(15,1)=5. gcoord(16,1)=0. gcoord(17,1)=1. gcoord(18,1)=2.5: gcoord(19,1)=3. gcoord(20,1)=5. gcoord(21,1)=0. gcoord(22,1)=1. gcoord(23,1)=2: gcoord(24,1)=3. ‘gcoord(25,1)=5. % | aan % input data for nodal connectivity for each element % nodes(igj) where i-> element no. and j-> connected nodes $$$ nodes(1,1)=1; nodes(1,; nodes(2,1)=2; nodes(2, nodes(3,1)=3; nodes(3;; nodes(4,1)=4; nodes(4,: nodes(19,1)=13; nodes(19,2)=14; nodes(19,3)=19; 109110 Laplace’s and Poisson’s Equations Chapter 5 nodes(20,1)=14; nodes(20,2)=15; nodes(20,3)=! nodes(21,1)=11; nodes(21,2)=17; nodes(21,3)=16; 5 nodes(22,3) }=19; nodes(23,3)=18; 10; nodes( 24,3): 7; nodes(25,3)=22; 7; nodes(26,2)=18; nodes(26,3) nodes(27,; odes(27,3): nodes(28,2)=20; nodes(28;; 6; nodes(29,2)=22; nodes(29,3)=21; =17; nodes(30,2)=23; nodes(30,3)=22; nodes(31,2)=24; nodes(31,3)=23; 9; nodes(32,2)=25; nodes(32,3)=24; % input data for boundary conditions (saan <= bedof(1)=1; % first node is constrained beval(1)=0; % whose described value is 0 bedof(2) % second node is constrained beval(2): % whose described value is 0 bedof(3)=3; % third node is constrained beval(3)=0; % whose described value is 0 bedof(4)=4; % 4th node is constrained beval(4)=0; % whose described value is 0 bedof(5) % Sth node is constrained beval(5): % whose described value is 0 bedof(6)=6; % 6th node is constrained beval(6)=0; % whose described value is 0 bedof(7)=11; % 11th node is constrained beval(7) % whose described value is 0 bedof(8)=16; % 16th node is constrained beval(8)=0; % whose described value is 0 bedof(9)=21; % 2st node is constrained beval(9): % whose described value is 0 bedof(10 % 22nd node is constrained beval(10)=38.2683; % whose described value is 38.2683 bedof(11 % 23rd node is constrained beval(11)=70.7107; % whose described value is 70.7107 bedof(12 % 24th node is constrained beval (12) bedof(13 beval(13) % %— % initialization of matrices and vectors ——————————— ff=zeros(sdof,1); % initialization of system force vector % whose described value is 92.3880 % 25th node is constrained % whose described value is 100Section 5.9 MATLAB Application to 2-D Steady State kk=zeros(sdof,sdof); % initialization of system matrix index=zeros(nnel*ndof,1); % initialization of index vector % _ % computation of element matrices and vectors and their assembly %——__— SSS for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ast connected node for (iel)-th element nd(2)=nodes(iel,2); % 2nd connected node for (iel)-th element nd(3)=nodes(iel,3); % 3rd connected node for (iel)-th element coord(nd(1),1); yl=geoord(nd(1),2); _-% coord values of Ist node x2=gcoord(nd(2),1); y2=gcoord(nd(2),2); _% coord values of 2nd node x3=gcoord(nd(3),1); y3=gcoord(nd(3),2); -% coord values of 3rd node % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=felp2dt3 (x1,y1,x2,y2,x3,y3); % compute element matrix % kk=feasmbl1 (kk,k,index); % assemble element matrices % end % A % apply boundary conditions %— —$_____ [kk, ff]=feaplyc2(kk,ff, bedof, beval); % ~%— % solve the matrix equation %— faol=kk\ ff; % go % analytical solution % for node x=gcoord(i,1); y=gcoord(i,2); esol(i)=100*sinh (0.31415927*y)*sin(0.31415927*x) /sinh(3.1415927); end % ——$—$ $$ $$ % print both exact and fem solutions it112 Laplace’s and Poisson’s Equations Chapter 5 function [kk]=feasmbll (kk,k,index) me — % Purpose: % Assembly of element matrices into the system matrix % % Synopsis: % (kk]=feasmbl1(kk,k,index) % % Variable Description: % kk - system matrix % k - element matri % index - d.o.f. vector associated with an element function [index]=feeldof(nd nnel,ndof) % Purpose: % Compute system dofs associated with each element. % % Synopsis: % {index]=feeldof(nd,nnel,ndof) % % Variable Description: % index - system dof vector associated with element ie! % nd - element node numbers whose system dofs are to be determined % nnel - number of nodes per element % ndof - number of dofs per node %Section 5.9 MATLAB Application to 2-D Steady State end end % function (kJ=felp2dt3(x1,y1,x2,y2,x3,y3) % % Purpose: % clement matrix for two-dimensional Laplace’s equation % using three-node linear triangular element % % Synopsis: % [k}=felp2dt3(x1,y1x2,y2,x3,y3) % % Variable Description: % - element stiffness matrix (size of 3x3) % x1, yl - x and y coordinate values of the first node of element % x2, y2 - x and y coordinate values of the second node of element % x3, y3 - x and y coordinate values of the third node of element % % % element matrix % A=0.5"(x2*y3-+x1 *y2+x3*y1-x2*y1-x1 *y3-x3*y2); % % area of the triangle K(1,1)=((x3-x2)*(x3-x2)+(y2-y3)*(y2-y3))/(4*A); K(1,2)=((x3-x2)*(x1-x3)+ (y2-y3)*(y3-y1))/(4*A); K (1,3) = ((x3-x2)*(x2-x1) + (y2-y3)*(y1-y2))/(4*A); (2,1) =k(1,2) 4 (2,2)=((x1-x3)*(x1-x3)+-(y3-y1)*(y3-y1))/(4*A); ke (2,3)=((x1-x3)*(x2-x1) +(y3-y1)*(y1-y2))/(4*A); k(3,1)=k(1,3); k(3,2)=k(2,3); K(3,3)=((x2-x1)*(x2-x1) + (y1-y2)*(y1-y2))/(4*A); % ‘The finite element and analytical solutions are compared below: store = dof # femsol exact 1.0000 0.0000 0.0000 2.0000 0.0000 0.0000 3.0000 0.0000 0.0000 4.0000 0.0000 0.0000 5.0000 0.0000 0.0000 6.0000 0.0000 0.0000 7.0000 3.6896, 2.8785, 8.0000 6.5689 5.3187114 Figure 5.9.2 Mesh With Bilinear Elements 9.0000 10.000 11.000 12.000 13.000 14.000 15,000 16.000 17.000 18.000 19.000 20.000 21,000 22,000 23.000 24,000 25.000 Laplace’s and Poisson’s Equations pele [|e oo 8.4046 9.0361 0.0000 10.621 17.312 21.626 23.124 0.0000 14.044 37.568 46.108 49.199 0.0000 38.268, 70.711 92.388 100.00 & Example 5.9.2 [elelele| 6.9492 7.5218 0.0000 7.6257 14.090 18.410 19.927 0.0000 17.324 32.010 41.823 45.269 0.0000 38.268 70.711 92.388 100.00 sielo Chapter 5 % at x=3.75 and y= % at x=5.00 and % at x=2.50 and y: % at x=3.75 and % at: x=5.00 and y=5. % at x=0.00 and % at x=1.25 and % at x=2.50 and y: % at. x=3.75 and % at x=5.00 and y: % at x=0.00 and % at x=1.25 and y= % at x=2.50 and % at x=3.75 and y=10. % at: x=5.00 and y=10. BBSaecaeeeeeees We solve the same problem as given in Example 5.9.1 using bilinear rectangular elements. The mesh is shown in Fig. 5.9.2 and the MATLAB programs are listed below. % % EX5.9.2.mSection 5.9 MATLAB Application to 2-D Steady State 115 % to solve the two-dimensional Laplace equation given as % uxx + yy =0,0 node no. and j-> x or y ———_Z—_~——Ee——— gcoord(1, gcoord(2, gcoord(3,1)=2.5; gcoord(3,2) gcoord(4,1)=3.75; gcoord(4,2)=0.0; 0; gcoord(5,2) .0; gcoord(1,2) .75; gcoord(14,2)=5.0; ; gcoord (15,2)Laplace’s and Poisson’s Equations Chapter 5 % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes 0; nodes(4,4)=9; nodes(5,4)=11; 3; nodes(6,4)=12; 4; nodes(74)=13; nodes(8,1)=9; nodes(8,2)=10; nodes(8,3)=15; nodes(8,4)=14; ; nodes(9,2)=12; nodes(9,3)=17; nodes(9,4)=16; }=12; nodes(10,2)=13; nodes(10,3)=18; nodes(10,4)= ; nodes(11,2)=14; nodes(11,3)=19; nodes(11,4) 14; nodes(12,2)=15; nodes(12,3)=20; nodes(12,4): ; nodes(13,2)=17; nodes(13,3)=22; nodes(13,4): 17; nodes(14,2)=18; nodes(14,3)=23; nodes(14,4) nodes(15,1)=18; nodes(15,2)=19; nodes(15,3)=24; nodes(15,4)=23; nodes(16, ; nodes(16,2)=20; nodes(16,3)=25; nodes(16,4)=24; % oe % input data for boundary conditions % first node is constrained % whose described value is 0 % second node is constrained % whose described value is 0 % third node is constrained % whose described value is 0 % Ath node is constrained % whose described value is 0 % Sth node is constrained % whose described value is 0 % 6th node is constrained % whose described value is 0 bedof(7)=11; % 11th node is constrainedSection 5.9 MATLAB Application to 2-D Steady State % whose described value is 0 % 16th node is constrained % whose described value is 0 % 2st node is constrained % whose described value is 0 bedof(10)=22; % 22nd node is constrained beval(10)=38.2683; % whose described value is 38.2683 % 23rd node is constrained % whose described value is 70.7107 bedof(12)=24; % 24th node is constrained beval(12)=92.3880; % whose described value is 92.3880 bedof(13)=25; % 25th node is constrained beval(1. 5 % whose described value is 100 % —E— % initialization of matrices and vectors a % ff=zeros(sdof,1); alization of system force vector kk=zeros(sdof,sdof); % initialization of system matrix index=zeros(nnel*ndof,1); % initialization of index vector % % % computation of element matrices and vectors and their assembly % loop for the total number of elements % loop for number of nodes per element ‘% extract connected node for (iel)-th element x(i)=gcoord(nd(i),1); % extract x value of the node y(i)=gcoord (nd (i),2); % extract y value of the node end % xleng = x(2)-x(1); % length of the element in x-axis yleng = y(4)-y(1); % length of the element in y-axis index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=felp2dr4(xleng,yleng); % compute element matrix % kk=feasmbl1(kk,k,index); % assemble element matrices % end % % % apply boundary conditions ges [kk, ff]=feaplyc2 (kk, ff, bedof, beval); % 117118 Laplace’s and Poisson’s Equations Chapter 5 %— % solve the matrix equation Ee feol=kk\ ft; % % % analytical solution %— for i=1:nnode x=gcoord(i,1); y=gcoord(i,2); esol(i)=100*sinh(0.31415927*y) *sin(0.31415927*x)/sinh(3.1415927); end % %— % print both exact and fem solutions % num=1:1:sdof; store=[num’ fsol esol’] % function [k]=felp2dr4(xleng,yleng) %— % Purpose: % element matrix for two-dimensional Laplace’s equation % using four-node bilinear rectangular element % % Synopsis: % [k]=felp2dr4(sleng,yleng) % % Variable Description: % k - element stiffness matrix (size of 4x4) % xleng - element size in the x-axis % yleng - element size in the y-axis (a % % element matrix % k(1,1)=(xleng*xleng+yleng*yleng) /(3*xleng*yleng); k(1,2)=(xleng*xleng-2*yleng*yleng) /(6*xleng*yleng); k(1,3)=-0.5*k(1,1); k(1,4)=(yleng* yleng-2*xleng*xleng)/(6*xleng*yleng); k(2,1)=k(1,2); k(2,2)=k(1,1)5 k(2,3)=k(1,4); k(2,4)=k(1,3); k(3,1)=k(1,3); k(3,2)=k(2,3); k(3,3)=k(1,1); k(3,4)=k(1,2); k(4,1)=k(1,4); k(4,2) % (2,4); k(4,3)=k(3,4); k(4,4)=k(1,1);Section 5.10 MATLAB Application to Axisymmetry 119 ‘The finite element solution is shown below. The same number of nodes were used for this case as that in the previous example. By comparing the two finite element solutions using either linear triangular elements or bilinear rectangular elements, we see that the rectangular elements produced more accurate solution in the present example. store = dof # femsol exact. 1.0000 0.0000 0.0000 % at x=0.00 and y=0.0 2.0000 0.0000 0.0000 3.0000 0.0000 0.0000 4.0000 0.0000 0.0000 5.0000 0.0000 0.0000 6.0000 0.0000 0.0000 7.0000 = 2.6888 2.8785 8.0000 4.9683 5.3187 9.0000 6.4914 69492 10.000 7.0263 «7.5218 11.000 0.0000 9.0000 12.000 7.2530 7.6257 13.000 13.402 14.090 14.000 17.510 18.410 15.000 18.953 19.927 16.000 0.0000 ~—0.0000 17.000 16.876 17.324 18.000 31.182 32.010 19.000 40.742 41.823 20,000 44,098 45.269 21.000 0.0000 0.0000 5 22.000 38.268 «38.268 % at x=1.25 and y=10. 23.000 = 70.711 70.711 % at x=2.50 and 24.000 92.388 «92.388 % at x=3.75 and 25.000 100.00 100.00 % at x: 5.10 MATLAB Application to Axisymmetric Analysis This section shows an example of an axisymmetric steady state problem using MATLAB programs. Linear triangular elements are used. # Example 5.10.1 An axisymmetric Laplace equation is solved using linear triangular elements. ‘The governing equation is given in Eq. (5.7.2) for a cylinder whose inside and outside radii are 4 and 6, and whose height is 1. The finite element mesh used for the present analysis is shown in Fig. 5.10.1. The boundary conditions are u = 100 at the inside of the cylinder and 34 = 20120 Laplace’s and Poisson’s Equations Chapter 5 ean ©| /® Ne Af Od 04 0 Od 04 Figure 5.10.1 An Infinite Cylinder Modeled With Symmetric Boundaries at the outside of the cylinder. Both top and bottom surfaces of the cylinder have 2% = 0, ie. insulated. Ten triangular elements with 12 nodes are used and the MATLAB programs are provided below. As seen in the main program, the constant flux at the outside surface is converted into the nodal flux at the outside surface. Each node takes a half of the total flux over the element which is 2nFql = 240m where 7=6, ¢=20 and [=1 as explained in Sec. 5.7. % % EX5.10.1.m % to solve the axisymmetric Laplace equation given as % ure + (ujr)/r + ujzz =0,4<1<6,0<2<1 % w(4,z) = 100, u,r(6,z) = 20 % u,x(2,0) = 0, u,2(r,1) = 0 % using linear triangular elements %(see Fig. 5.10.1 for the finite element mesh) % % Variable descriptions % k = element matrix % { = element vector % kk = system matrix % ff = system vector % gcoord = coordinate values of each node % nodes = nodal connectivity of each element % index = a vector containing system dofs associated with each element % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in bedof % % —$$$$ $$$ % input data for control parameters aSection 5.10 MATLAB Application to Axisymmetry 121 nel: % number of elements nnel=; % number of nodes per element ndot % number of dofs per node nnode=12; % total number of nodes in system sdof=nnode*ndof; % total system dofs % G6 % imput data for nodal coordinate values % gcoord(i,j) where i-> node no, and j-> x or y % scoord(1;1)=4.0; gooord(1,2)=0.; gcoord(2,1)=4.0; gcoord(2,2)=1.05 gcoord(3,1)=4.4; gcoord(3,2)=0. gcoord(5, gcoord(7,1 gcoord(9, .6; gcoord (10,2)=1.0; gcoord(11,1)=6.0; gcoord(11,2)=0.0; gcoord(12,1)=6.0; gcoord(12,2)=1.0; % % % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes rr nodes(8,3) =12; nodes(9,3)=1 11; nodes(10,3) % input data for boundary conditions % bedof(1 % first node is constrained beval(1)=100; % whose described value is 100 % second node is constrained beval(2)=100; % whose described value is 100 % —$ % initialization of matrices and vectors — ff=zeros(sdof,1); % initialization of system force vector kk=zeros(sdof,sdof); % initialization of system matrix index=zeros(nnel*ndof,1); % initialization of index vector % pi=d*atan(1); % define pi122 Laplace’s and Poisson’s Equations #€(11)=120* pi; f€(12)=120* pi; % % % for iel=1:nel % nd(1)=nodes(iel,1); nd(2)=nodestiel,2); nd(3)=nodes(iel,3); % nodal flux at the outside boundary % nodal flux at the outsdie boundary % computation of element matrices and vectors and their assembly % loop for the total number of elements % Ast connected node for (iel)-th element % 2nd connected node for (iel)-th element % 3rd connected node for (iel)-th element coord(nd(1),1); 1=gcoord(nd(1),2); % coordinate of Ist node coord(nd(2),1); 22=gcoord(nd(2),2); % coordinate of 2nd node 13=gcoord(nd(3),1); 23=gcoord(nd(3),2); % coordinate of 3rd node % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=felpaxt3(rl,z1,12,22,x3,23); % compute element matrix % Kk=feasmbl1 (kk,k,index); % assemble element matrices % end % ices oc % apply boundary conditions %&—________ {kk,ff]=feaplyc2 (kk, ff, bedof,bcval); % ———_____——_—— % solve the matrix equation a faol=kk\ff; % %e % analytical solution % for inode r=gcoord(i,1); z=gcoord(i,2); 00-6*20*1og(4)+6*20*log(r); esol (i)= end % —— % print both exact and fem solutions (ee num=1:1:sdof; store=[num’ fsol esol’] % % Chapter 5Section 5.10 MATLAB Application to Axisymmetry function (k]=felpaxt3(r1,z1,12,22,13,23) % % Purpose: % element matrix for axisymmetric Laplace equation % using three-node linear triangular element % % Synopsis: % [k)=felpaxt3(r1,21,12,22,13,23) % % Variable Description: % k - element stiffness matrix (size of 3x3) % 11, 21 - x and z coordinate values of the first node of element % 12, 22 - x and z coordinate values of the second node of element % 13, 23 - x and z coordinate values of the third node of element % % % element matrix % A=0.5*(12*23-+11*22-+13*21-12*2 1-11 *23-13*22); twopirc=8*atan(1)*r0; (1,1) = ( (r3-r2)* (13-12)-+(22-23)*(22-23))/(4*A); 4 (1,2}=({(r3-12)*(r1-13)+(22-23)*(23-21))/(4"A); k(1,3)=((r3-r2)*(r2-11) +(22-23)*(21-22))/(4"A); k(2,1)=k(1,2); k(2,2)=((r1-r3)" (r1-13)+ (23-21)"(23-21))/(4*A); K(2,3)=((r1-18)* (12-1) +(23-21)*(21-22))/(4*A); K(3,1}=k(1,3); k(3,2)=k(2,3); 1e(3,3)=((x2-11)* (r2-11)+(21-22)*(21-22))/(4*A); k=twopirc*k; % %e % area of the triangle 1+12413)/3; % x coordinate value of the element centroid ‘The results are store = dof # femsol — exact 1.0000 100.000 100.000 2.0000 100.000 100.000 3.0000 = 111.413 111.437 4.0000 111.444 111.437 5.0000 121.839 121.879 6.0000 121.889 121.879 7.0000 = 131.427 131.484 8.0000 = 131.501 :131.484 9.0000 140.290 140.377 10.000 140.417 140.377 11.000 148.510 148.656 % at r=4.0 and z=0.0 % at r=4.0 and 2=1.0 4 and 4 and 8 and 8 and eS Se5o5558 123124 Laplace’s and Poisson’s Equations Chapter 5 Figure 5.11.1 Mesh With Triangular Element 12.000 148.749 148.656 % at r=6.0 and 2=1.0 5.11 MATLAB Application to Transient Analysis Examples are given for some transient analyses using MATLAB programs. Forward difference, backward difference and Crank-Nicolson techniques are used. @ Example 5.11.1 The transient Laplace equation as described below is solved using the forward difference technique for time integration. The differential equation is ou Ou Pu Oa? t By? over a rectangular domain defined by 0 < 2 <5 and 0 < y < 2. The whole domain has the initial value of u = 0, and suddenly the left and right boundaries (i.e. edges with x=0 and x=5) are maintained at u = 100. On the other hand, the top and bottom boundaries (i.e. edges with y=0 and y=2) are insulated (i.e. Ea = 0). We want to find the solution as a function of time. A finite element (6.11.1) mesh of the domain is shown in Fig. 5.11.1 using 16 linear triangular elements. The finite element analysis program is shown below. %- % EXS.11.1.m % to solve the transient two-dimensional Laplace’s equation Hut =uxx + ayy, Onode no. and j->x or y126 Laplace’s and Poisson’s Equations gcoord(8,1)=2.5; gcoord(8,2)=1.0; gcoord(11,1)=0.0; gcoord(11,2)=2. geoord(12,1 gcoord(13,1): gcoord(14,1): gcoord(15,1): % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes — eee nodes(2,1) nodes(3,1)=: nodes(5,1)=1; nodes(6,1): nodes(7,1): nodes(8,1) nodes(9,1) nodes(10,1)=7; nodes(10,2)= nodes(11,1)=8; nodes(11,2’ nodes(13,1)=6; nodes(13,2' nodes(14,1)=7; nodes(14,2 nodes(15,1)=8; nodes(15,2) nodes(16,1)=9; nodes(16,2 % — —_—————— % input data for boundary conditions %— bedof(1) Chapter 5 4 % Ist node is constrained beval(1)=100; % whose described value is 100 bedof(2) % 5th node is constrained beval(2)=100; % whose described value is 100 % 6th node is constrained % whose described value is 100 % 10th node is constrained % whose described value is 100 % 11th node is constrained % whose described value is 100 bedof(6)=15; % 15th node is constrained beval(6)=. % — % initialization of matrices and vectors 10; % whose described value is 100Section 5.11 MATLAB Application to Transient Analysis %—— - ff=zeros(sdof,1); % initialization of system vector fn=zeros(sdof,1); % initialization of effective system vector fsol=zeros(sdof,1); % solution vector ros(2,ntime+1); % vector containing time history solution kk=zeros(sdof,sdof); % initialization of system matrix mm-=zeros(sdof,sdof); % initialization of system matrix index=zeros(nnel*ndof,1); % initialization of index vector % 6 % computation of element matrices and vectors and their assembly $$ for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ist connected node for (iel)-th element nd(2)=nodes(iel,2); % 2nd connected node for (iel)-th element nd(3)=nodes(iel,3); % 3rd connected node for (iel)-th element x1=gcoord(nd(1),1); yl=gcoord(nd(1),2); _% coord values of Ist node coord(nd(2),1); y2=gcoord(nd(2),2); % coord values of 2nd node coord(nd(3),1); y3=gcoord(nd(3),2); _% coord values of 3rd node index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=felp2dt3(x1,y1,x2,y2,x3,y3); % compute element matrix m=felpt 2t3(x1,y1,x2,y2,x3,y3); % compute element matrix % kk=feasmbl1(kk,k,index); % assemble element matrices mm=feasmbl1(mm,m index); % assemble element matrices % end % % % loop for time integration % initial condition sol(1,1)=fs0l(8); % store time history solution for node no. 8 sol(2,1)=fs0l(9); % store time history solution for node no. 9 a it=1:ntime % start loop for time integration ieee ae % compute effective column vector net ae a % apply boundary condition % fsol=mm\fn; % solve the matrix equation 127128 Laplace’s and Poisson’s Equations Chapter 5 % sol(1,it+1)=fsol(8); % store time history solution for node no. 8 sol(2,it+1)=fsol(9); ‘% store time history solution for node no. 9 % end % —————___———. % plot the solution at nodes 8 and 9 %—___— = time=0:deltt:ntime*deltt; plot(time,sol(1,:),",time,sol(2,:),"~"); xlabel(’Time’) ylabel(’Solution at nodes’) % %: function (m]=felpt2t3(x1,y1,x2,y2,x3,y3) %— % Purpose: % element matrix for transient term of two-dimensional % Laplace’s equation using linear triangular element % % Synopsis: % (m)=felpt2t3(x1,y1,x2,y2,x3,y3) % % Variable Description: % m - element stiffness matrix (size of 3x3) % x1, yl - x and y coordinate values of the first node of element % x2, y2 - x and y coordinate values of the second node of element % x3, y3 - x and y coordinate values of the third node of element %— Secale ela cts eects ee cee CeCe eeC eee EREUEEE % % element matrix % A=0.5*(x2*y3-4-x1 *y2-+x3*y1-x2*yl-x1*y3-x3*y2); % area of the triangle % m = (A/12)* [211 121; 112); %: The finite element solutions are plotted in Fig. 5.11.2 and Fig. 5.11.3. Time- history of nodes 8 and 9 in Fig. 5.11.1 is plotted in both figures. While At = 0.1 was used for Fig. 5.11.2, At = 0.12 was used for Fig. 5.11.3. As noticed, the finite element solution is unstable when At = 0.12 is used because the forward difference technique is conditionally stable. tSection 5.11 MATLAB Application to Transient Analysis 129 Figure 5.11.2 Finite Element Solution With At=0.1 Figure 5.11.3 Finite Element Solution With At=0.12 & Example 5.11.2 The same example as Example 5.11.1 is solved using bilinear rectangular elements. The mesh is the same as that shown in Fig. 5.11.1 except that 8 rectangular elements are used instead of 16 triangular elements. %— % EX5.11.2.m % to solve the transient two-dimensional Laplace’s equation Hut =uxx + ayy, 0 —<_—————_—____—__—— % number of elements % number of nodes per element % number of dofs per node nnode=15; % total number of nodes in system sdof=nnode*ndof; % total system dofs delt % time step size for transient analysis stime=0.0; % initial time ime=10; % termination time =fix((ftime-stime)/deltt); % number of time increment % input data for nodal coordinate values % gcoord(i,j) where i->node no. and j->x or y gcoord(7,1)=1.25; gcoord(7,2)=1.0; coord (8,2)=1.0;Section 5.11 MATLAB Application to Transient Analysis gcoord(12,1)=1.25; gcoord(i2, gcoord(13,1)=2.5; gcoord(13,2): gcoord(14,1)=3.75; gcoord(14,2)=2. gcoord(15,1)=5.0; gcoord(15,2)=2.0; % (a % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes % initialization of matrices and vectors % ff=ner08(sdof,1); in=zeros(sdof,1); fsol=zeros(sdof,1); sol=zeros(2,ntime+1); mm=zeros(sdof,sdof); index=zeros(nnel*ndof,1); % %- nodes(5,3)= nodes(6,3)=13; nodes(6,4)=12; nodes(7,3)=14; nodes(7,4)=13; 2; nodes(5,4)=11; % st node is constrained % whose described value is 100 % Sth node is constrained % whose described value is 100 % 6th node is constrained % whose described value is 100 % 10th node is constrained % whose described value is 100 % 11th node is constrained % whose described value is 100 % 15th node is constrained %- whose described value is 100 % initialization of system vector % initialization of effective system vector % solution vector % vector containing time history solution % initialization of system matrix % initialization of system matrix % initialization of index vector % computation of element matrices and vectors and their assembly —— 131Laplace’s and Poisson’s Equations Chapter 5 for iel=1:nel % nd(1)=nodes(iel,1); nd(2)=nodes(iel,2); nd(3)=nodes(iel,3); nd(4)=nodes(iel,3); % loop for the total number of elements % Ist connected node for (iel)-th element % 2nd connected node for (iel)-th element % 3rd connected node for (iel)-th element % 4th connected node for (iel)-th element coord (nd(1),1); y xleng=x2-x1; yleng=y4-y1; % index=feeldof(nd,nnel,ndof);% k=felp2dr4(xleng,yleng); m=felpt2r4(xleng,yleng); % kk=feasmbl1 (kk,k,index); mm=feasmbl1(mm,m,index); % end % % % loop for time integration % % for in=1:sdof fsol(in)=0.0; end % sol(1,1)=fs0l(8); sol(2,1)=f80l(9); % in=deltt*f+(mm-deltt*kk)*fsol; % {mm, fn]=feaplyc2(mm, fn, bedof,beval); % fsol=mm\fn; % sol(1,it-+1 sol(2,it-+1 % end % 7 % analytical solution at node 8 coord(nd(1),2); % coord values of Ist node scoord(nd(2), coord(nd(2),2); % coord values of 2nd node =gcoord(nd(3),2); % coord values of 3rd node coord(nd(4),2); % coord values of 4th node % element size in x-axis % element size in y-axis % time-independent element matrix % transient element matrix % assemble element matrices % assemble element matrices % initial condition % store time history solution for node no. 8 % store time history solution for node no. 9 % start loop for time integration % compute effective column vector % apply boundary condition % solve the matrix equation % store time history solution for node no. 8 % store time history solution for node no. 9Section 5.11 MATLAB Application to Transient Analysis 133 %—__— pi=4*atan(1); esol=zeros(1,ntime+1); x=2.5; x1=5; esol (ii)=esol(ii)+(1/i)*exp(-iti*pi* pi*ti/(x1*xl))*sin(i*pi*xx/xl); end end esol=100-(100*4/pi)*esol; % —— % plot fem and exact solutions at node 8 YX $$$ ——————_—_———— time=0:deltt:ntime*deltt; plot(time,sol(1,:),"*’,time,esol,’-’); xlabel(’Time’) ylabel(’Solution at nodes’) % % function {m]=felpt2r4(xleng,yleng) %— % Purpose: % element matrix of transient term for two-dimensional Laplace’s % equation using four-node bilinear rectangular element % % Synopsis: % (m]=felpt2r4(xleng,yleng) % % Variable Description: % m - element stiffness matrix (size of 4x4) % xleng - clement size in the x-axis % yleng - element size in the y-axis % % % element matrix % m=(xleng*yleng/36)*[4 2 1 2; 2421; 1242; 2124), % ‘The finite clement solution obtained using rectangular elements is comparable to that obtained using triangular elements. Figure 5.11.4 compares the finite134 Laplace’s and Poisson’s Equations Chapter 5 Figure 5.11.5 Time History of Nodes 8 and 9 clement solution to the exact solution at node 8 (see Fig. 5.11.1). The finite element solution approaches the steady state solution slower than the exact solution. This is due to the very crude mesh in the 2: t irection. Example 5.11.3 The present example solves the same problem as that in Example 5.11.1 using the backward difference technique for time integration. Because this technique is unconditionally stable, we use a time step size At = 0.4 which exceeds the critical time step size for the forward difference technique. Figure 5.11.5 shows the time-history of nodes 8 and 9. SS % EX5.11.3.mSection 5.11 MATLAB Application to Transient Analysis 135 % to solve the transient two-dimensional Laplace’s equation Hut =uxxtuyy,0node no. and j->x or y gcoord(1,1)=0.0; gcoord(1,2): gcoord(2,1)=1.25; gcoord(2, gcoord(3,1)=2.5; gcoord(3,2) 0;136 Laplace’s and Poisson’s Equations Chapter 5 geoord(4,1)=3.75; gcoord(4,2)=0. gcoord(5,1)=5.0; gcoord(5,2)=0.0; gcoord(7,1)=1.25; gcoord(7,2)=1.0; 5; gcoord(8,2)=1.0; gcoord(12,1)=1.25; gooord(12,2)=2.0; gcoord(13,1)=2.5; gcoord(13,2)=2. % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes %—.—— nodes(1,1)=1; nodes(1,2)=: nodes(2,1)=2; nodes(2,2): nodes(3,1)=3; nodes(3,2): nodes(5,1)=1; nodes(5,2): nodes(6,1)=2; nodes(6,2)= nodes(12,1 nodes(13,1)=6; nodes(13,2)=12; nodes(13,3)=11; nodes(16,1)=9; nodes(16,2)=15; nodes(16,3)=14; % §——$ % input data for boundary conditions ———————— bedof(1): % Ist node is constrained beval(1)=100; % whose described value is 100 bedof(2): % Sth node is constrained beval(2)=100; % whose described value is 100 bedof(3) % 6th node is constrained beval(3)=100; % whose described value is 100 bedof(4)=10; % 10th node is constrained beval(4)=100; % whose described value is 100 bedof(5)= % 11th node is constrained beval(S)=100; % whose described value is 100 bedof(6)=15; % 15th node is constrainedSection 5.11 MATLAB Application to Transient Analysis beval(6)=100; - % whose described value is 100 % —— $$ $$$ % initialization of matrices and vectors % ee ff=zeros(sdof,1); % initialization of system vector fn=zeros(sdof,1); % initialization of effective system vector fsol=zeros(sdof,1); % solution vector jeros(2,ntime+1); % vector containing time history solution ros(sdof,sdof); % initialization of system matrix mm=zeros(sdof,sdof); % initialization of system matrix index=zeros(nnel*ndof,1); % initialization of index vector % —————————— % loop for the total number of elements % Ist connected node for (iel)-th clement % 2nd connected node for (iel)-th element % 3rd connected node for (iel)-th element oord(nd(1),1); y1=gcoord(nd(1); % coord values of Ist node x2=gcoord(nd(2),1); y2=gcoord (nd(2), % coord values of 2nd node oord(nd(3),1); y3=gcoord(nd(3),2); _% coord values of 3rd node index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=felp2dt3(x1,y1,x2,y2,x3,y3); % compute element matrix m=felpt2t3(x1,y1,x2,y2,x3,y3); % compute element matrix % kk=feasmbl1(kk,k,index); % assemble element matrices mm=feasmbl1(mm,m, index); % assemble element matrices % end % Cu % loop for time integration % initial condition % sol contains time-history solution of node 8 % sol contains time-history solution of node 9 kk=mm4deltt*kk; % for it: 137138 Laplace’s and Poisson’s Equations Chapter 5 3 insulated 0.02m pe —_—_ adem Figure 5.11.6 Finite Element Mesh % fn=deltt*ff-+mm*fsol; % compute effective column vector % [kk,fn]=feaplyc2(kk,fn, bedof, beval); % apply boundary condition % fsol=kk\ fn; % solve the matrix equation % sol(1,it+1)=fs0l(8); % sol contains time-history solution of node 8 sol(2,it-+1)=fs0l(9); % sol contains time-history solution of node 9 % end % — ee % plot the solution at nodes 8 and 9 ———————————— time=0:deltt:ntime*deltt; plot(time,sol(1,:),*°,time,sol(2,:),-”); xlabel(’Time’) ylabel(’Solution at nodes’) % % @ Example 5.11.4 A plate of size 0.02 m by 0.01 m, whose heat conduction coefficient is k = 0.3 W/mC, is initially at a temperature of 300 C. While its left and right sides are maintained at the same temperature of 300 C, the bottom side is insulated and the top side is subjected to heat convection with convection coefficient of he = 100 W/m? C and the ambient temperature of 50 C. The material has also density p=1600 Kg/m? and specific heat c=0.8 J/KgC. TheSection 5.11 MATLAB Application to Transient Analysis 139 a a Figure 5.11.7 Time History Plot MATLAB program using the backward difference technique is listed below and the mesh is shown in Fig. 5.11.6. The time-history of the solution of node 8 is given in Fig. 5.11.7. % % EX5.11.4.m % to solve the transient two-dimensional Laplace’s equation % atu,t = uxx + wyy,0 node no. and j-> x or y ————————— 0.0; gcoord(1,2): .005; gcoord(2,2) gcoord(4, gcoord(5,1)=0.020; gcoord(5,2)=0.0; gcoord(6,1)=0.0; gcoord(6,2)=0.005; gcoord(7,1)=0.005; gcoord(7,2) gcoord(8,1)=0.010; gcoord(8,2) gcoord(9, gcoord(10,1 gcoord(11,1 gcoord(12,1)=0.005; gcoord(12,2)=0.01; gcoord(13,1)=0. %&— % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes % ee nodes(1,1)=1; nodes(2,1Section 5.11 MATLAB Application to Transient Analysis nodes(4,1) nodes(5,1): nodes(6,1) nodes(4,3)=10; nodes(4, nodes(5,3)=12; nodes(5,4)= % Ast node is constrained % whose described value is 300 % 5th node is constrained % whose described value is 300 % 6th node is constrained % whose described value is 300 % 10th node is constrained % whose described value is 300 % 11th node is constrained % whose described value is 300 % 15th node is constrained % whose described value is 300 % input for flux boundary conditions % nfix(ij) where i-> element no. and j-> two side nodes % nodes on 1st element side with flux % nodes on 2nd element side with flux % nodes on 3rd element side with flux nfix(4,1)=14; nfix(4,2)=15; % nodes on 4th element side with flux % b=100; c=50; % %——______— % initialization of matrices and vectors Ca ff=zeros(sdof,1); % system vector fn=zeros(sdof,1); % effective system vector fsol=zeros(sdof,1); % solution vector sol=zeros(1,ntime+1); ‘% time-history of a selected node kk=zeros(sdof,sdof); % of system matrix mm=zeros(sdof,sdof); % system matrix index=zeros(nnel*ndof,1); % index vector fl=zeros(nnels*ndof,1); % clement flux vector k1=zeros(nnels*ndof,nnels*ndof); % flux matrix index1=zeros(nnels*ndof,1); % flux index vector % Ge % computation of element matrices and vectors and their assembly 141142 Laplace’s and Poisson’s Equations Chapter 5 % for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ast connected node for (iel)-th element % 2nd connected node for (iel)-th element % 3rd connected node for (iel)-th element % 4th connected node for (iel)-th element =gcoord(nd(1),2); % coord values of Ist node oord(nd(2),1); y2=gcoord(nd(2),2); % coord values of 2nd node % coord values of 3rd node % coord values of 4th node % element size in x-axis % clement size in y-axis index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=felp2dr4(xleng,yleng); % compute element matrix m=a*felpt2r4(xleng,yleng); % compute element matrix % kk=feasmbl1 (kk,k,index); % assemble element mat mm=feasmbl1(mm,m,index); % assemble element matrices % end % eee dela aeeaee eee eee eatec eC eee eueas % additional computation due to flux boundary condition %— ———————— for ifx=I:nf % nds(1)=nflx(ifx,1); % node with flux BC for (ifx)-th element nds(2)=nflx(ifx,2); % node with flux BC for (ifx)-th element x1=gcoord(nds(1),1); y1=gcoord(nds(1),2); % nodal coordinate cord (nds(2),1); y2=gcoord(nds(2),2); % nodal coordinate eleng=sqrt((x2-x1)*(x2-x1)+(y2-y1)*(y2-y1));_-% element side length oe % find related system dots ee % compute element matrix due to flux fl=b*c*fefll(0,eleng); % compute element vector due to flux oe jindex1); % assembly % end % % % loop for time integrationSection 5.11 MATLAB Application to Transient Analysis 143 fsol(in)=300.0; % initial condition end % sol(1)=fs0l(8); % sol contains time-history solution at node 8 kk=mm-+deltt*kk; % effective system matrix % for it=1:ntime % fn=deltt*ff+mm*fsol; % compute effective column vector % [kk,fn]=feaplyc2(kk, fn, bcdof, beval); % apply boundary condition % fsol=kk\ fn; % solve the matrix equation % sol(it-+-1)=fsol(8); % sol contains time-history solution at node 8 % end % eo % plot the solution at node 8 %— time=0:deltt:ntime*deltt; plot(time,sol); xlabel(’Time’) ylabel(’Solution at the center’) % %- function [k]=feflxl2(eleng) % % Purpose: % element matrix for Cauchy-type boundary such as du/dn=a(u-b) % using linear element where a and b are known constants. nh % Synopsis: % [k]=feflxl2(eleng) % % Variable Description: % k - element vector (size of 2x2) % eleng - length of element side with given flux % * % element matrix % k=(eleng/6)*[ 21;144 Laplace's and Poisson’s Equations Chapter 5 ‘sen at ca Seeeee ss 8 OTe at Figure 5.11.8 Transient Finite Element Solution 123; % % & Example 5.11.5 We use the Crank-Nicolson technique to solve the following problem. yp? o0s3t = ea oe (5.11.2) ‘The problem domain is the same as that shown in Fig. 5.9.1 and the boundary conditions are the same as those described in Example 5.9.1. The initial condition is 100. The transient solution is plotted in Fig. 5.11.8. The steady state solution is that obtained in Example 5.9.1. As a result, the present solution must approach the steady state solution. The program is given below. %— % EX5.11.5.m % to solve the two-dimensional Laplace’s equation given as % 0.04*u,t = uxx + wyy ,0 node no. and j-> x or y gcoord(9,1)=3.75; gcoord(9,2)=2.5; gcoord(10,1)=5.0; gcoord(10,2)=2.5; gcoord(11,1)=0. gcoord(12,1)=1.25; gcoord(12,2) gcoord(13,1)=2.5; gcoord(13,2)=5.0; gcoord(14,1)=3.75; gcoord(14,2)=5.0; gcoord(15,1) . gcoord(16,1)=0.0; gcoord(16,2)=7.5; gcoord(17,1)=1.25; gcoord(17,2) gcoord(18,1)=2.5; gcoord(18,2)=7.Laplace’s and Poisson’s Equations Chapter 5 gcoord(19,1)=3.75; gcoord(19,2)=7.: 5; gcoord(20,1)= .75; gcoord(24,2)=10, gcoord(25,1)=5.0; gcoord(25,2)= % a % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes —_—_&_——_—__—_— nodes(2,1 nodes(4,1)=4; nodes(4 2)=5; nodes(4,3)=10; nodes(4,4)=9; nodes(5,1)=6; nodes(5,2 2; nodes(5,4)=11; nodes(6,1)=7; nodes(6,2)=8; nodes(7,1)=8; nodes(7,2 nodes(8,1)=9; nodes(8,2 nodes(9,1 nodes(13,1)=16; nodes(13,2)=1 nodes(14,1)=17; nodes(14,2): nodes(15,1)=18; nodes(15,2) nodes(16,1)=19; nodes(16,2)=: % See eeeee cae e seeare cece % input data for boundary conditions _ bedof(1 4; nodes(15,4 5; nodes(16,4) % Ast node is constrained % whose described value is 0 % 2nd node is constrained % whose described value is 0 ‘% 3rd node is constrained % whose described value is 0 % 4th node is constrained % whose described value is 0 % 5th node is constrained ‘% whose described value is 0 % 6th node is constrained % whose described value is 0 % 11th node is constrained % whose described value is 0 % 16th node is constrained % whose described value is 0Section 5.11 MATLAB Application to Transient Analysis bedof(9)=23; % 2st node is constrained % whose described value is 0 % 22nd node is constrained % whose described value is 38.2683 % 23rd node is constrained % whose described value is 70.7107 % 24th node is constrained % whose described value is 92.3880 % 25th node is constrained % whose described value is 100 % initialization of matrices and vectors eros(sdof,1); % system vector In=seros(edef, 1); % effective system vector % solution vector seen ntime+1); % time-history solution kk=zeros(sdof,sdof); % initialization of system matrix mm-=zeros(sdof,sdof); % initialization of system matrix kn=zeros(sdof,sdof); % effective system matrix index=zeros(nnel*ndof,1); % initialization of index vector % Gs % computation of element matrices and vectors and their assembly ——_ for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ist connected node for (iel)-th element nd(2)=nodes(iel,2); % 2nd connected node for (iel)-th element nd(3)=nodes(iel,3); % 8rd connected node for (iel)-th element nd(4)=nodes(iel,4); % 4th connected node for (iel)-th element x1=gcoord(nd(1),1); y’ x2=gcoord(nd(2),1); y x3=gcoord(nd(3),1); y3: x4=gcoord(nd(4),1); y: oord(nd(1),2); % coord values of Ist node coord(nd(2),2); % coord values of 2nd node coord(nd(3),2); % coord values of 3rd node ‘oord(nd(4),2); % coord values of 4th node xleng=x2-x1; % element size in x-axis yleng=y4-y1; % element size in y-axis index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=felp2dr4(xleng,yleng); % compute element matrix m=a*felpt2r4(xleng,yleng); % compute element matrix % asmbl1 (kk,k,index); % assemble element matrices =feasmbl1 (mm,m,index); % assemble element matrices 147148 Laplace’s and Poisson’s Equations Chapter 5 % a % loop for time integration a for in=1:sdof fsol(in)=100.0; % initial condition end % sol(1)=fsol(13); % sol contains time-history solution at node 13 % kn=2*mm-+deltt*kk; % compute effective system matrix fn=deltt*ff+(2*mm-deltt*kk)*fsol; % compute effective vector % fin, fn]=feaplyc2(kn, fn, bedof, beval); % apply boundary condition % fsol=kn \fn; % solve the matrix equation % sol(it+1)=fsol(13); % sol contains time-history at node 13 % end % % % plot the solution at node 13 G——$ $$$ time=0:deltt:ntime*deltt; plot(time,sol); xlabel("Time’) ylabel(’Solution at the center’) % %— 5.12 MATLAB Application to 3-D Steady State Analysis 4 Example 5.12.1 A pyramid shape of three-dimensional domain as seen in Fig. 5.12.1 is analyzed for the Laplace equation. The bottom face of the pyramid has specified nodal variables as given in the figure while the side faces ou have no flux (i.e. 55 = 0), Four tetrahedral elements are used for the present three-dimensional analysis. The MATLAB program is also listed.Section 5.12 MATLAB Application to 3-D Steady State 149 (0.5,0.5,1) ug=100 Figure 5.12.1 A Pyramid With Four Tetrahedral Elements %— % EX5.12.1.m % to solve the three-dimensional Laplace equation % for a pyramid shape of domain % using four-node tetrahedral elements. % Bottom face has essential boundary condition and the % side faces are insulated. %(see Fig. 5.12.1 for the finite element mesh) % % Variable descriptions % k = element matrix % { = element vector % kk = system matrix % ff = system vector % gcoord = coordinate values of each node % nodes = nodal connectivity of each element % index = a vector containing system dofs associated with each element % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with . the dots in bedof % number of elements % number of nodes per element % number of dofs per node % total number of nodes in system sdof=nnode*ndof; % total system dofs %150 Laplace’s and Poisson’s Equations Chapter 5 $$$ $ % input data for nodal coordinate values % gcoord(i,j) where i-> node no, and j-> x or y EE % input data for nodal connectivity for each element % nodes(ij) where i-> element no. and j-> connected nodes — nodes(1,1)=4; nodes(1,2)=1; nodes(1,3)=3; nodes(1,4)=« nodes(2,1)=1; nodes(2,2)=2; nodes(2,3)=3; nodes(2,4) nodes(3,1)=2; nodes(3,2)=5; nodes(3,3)=3; nodes(3,4) nodes(4,1)=5; nodes(4,2)=4; nodes(4,3)=3; nodes(4,4): % ————— % input data for boundary conditions ee bedof(1): % Ast node is constrained beval(1) % whose described value is 0 bedof(2): % 2nd node is constrained beval(2): % whose described value is 20 bedof(3): % 4th node is constrained beval(3)=50; % whose described value is 50 bedof(4)=5; % 5th node is constrained beval(4)=100; % whose described value is 100 h %——— $$ __—_—_—_ % initialization of matrices and vectors %— = ff=zeros(sdof,1); % system vector kk=zeros(sdof,sdof); % system matrix index=zeros(nnel*ndof,1); % index vector % % computation of element matrices and vectors and their assembly % loop for the total number of elements % 1st connected node for (iel)-th element % 2nd connected node for (iel)-th element % 3rd connected node for (iel)-th element % 4th connected node for (iel)-th elementSection 5.12 MATLAB Application to 3-D Steady State 151 x(1)=gcoord(nd(1),1); y(1)=gcoord(nd(1),2); 2(1)=gcoord(nd(1),3); % coordinate of Ist node x(2)=gcoord(nd(2),1); y(2)=gcoord(nd(2),2); 2(2)=gcoord(nd(2),3); % coordinate of 2nd node x(3)=gcoord(nd(3),1); y(3)=gcoord(nd(3),2); 2(3)=gcoord(nd(3),3); % coordinate of 3rd node x(4)=gcoord(nd(4),1); y(4)=gcoord(nd(4),2); 2(4)=gcoord(nd(4),3); % coordinate of 4th node wh index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=felp3dt4(x,y,2); % compute element matrix kk=feasmbl! (kk,k,index); % assemble element matrices % end % qe % apply boundary conditions %———$______—- [kk, ff] =feaplyc2(kk,ff,bedof, beval); % %— % solve the matrix equation %— faol=kk\ ft; %—_—— % print both exact and fem solutions a num: sdof; store=[num’ fsol] % %— function [k]=felp3dt4(x,y,z) %— Se % Purpose: % element matrix for three-dimensional Laplace’s equation % using four-node tetrahedral element % % Synopsis: % [k}=felp3dta(x,y,z) % % Variable Description: % k - element matrix (size of 4x4)Laplace’s and Poisson’s Equations Chapter 5 % x - x coordinate values of the four nodes % y - y coordinate values of the four nodes % 2 - & coordinate values of the four nodes %e % xbar= [1 x(1) y(1) 2(1); 1 x(2) y(2) 2(2)5 1 x(3) y(3) 2(3)5 1 x(4) y(4) 2(4) Js xinv = inv(xbar); vol = (1/6)*det(xbar); % % element matrix % K(1,1)=xinv(2,1)*xinv(2,1)-+xinv(3,1)*xinv(3,1)+xinv(4,1)*xinv(4,1); k(1,2)=xinv(2,1)*xinv(2,2)+xinv(3,1)*xinv(3,2)+xinv(4,1)*xinv(4,2); k(1,3)=xinv(2,1)*xinv(2,3)-+xinv(3,1)*xinv(3,3)+xinv(4,1)*xinv(4,3); k(1,4)=xinv(2,1)*xinv(2,4)+xinv(3,1)*xinv(3,4)+xinv(4,1)*xinv(4,4); k(2,1}=k(1,2); k(2,2)=xinv(2,2)*xinv(2,2)-+xinv(3,2)*xinv(3,2)+xinv(4,2)*xinv(4,2); k(2,3)=xinv(2,2)*xinv(2,3)+xinv(3,2)*xinv(3,3)+xinv(4,2)*xinv(4,3); k(2,4)=xinv(2,2)*xinv(2,4)+xinv(3,2)*xinv(3,4)+xinv(4,2)*xinv(4,4); k(3,1)=k(1,3); k(3,2)=k(2,3); k(3,3)=xinv(2,3)*xinv(2,3)4+xinv(3,3)*xinv(3,3)+xinv(4,3)*xinv(4,3); K(3,4 v(2,3)*xinv(2,4)-+xinv(3,3)*xinv(3,4)-+xinv(4,3)*xinv(4,4); k(4,1 1,4); K(4,2)=k(2,4); =k(3,4); inv(2,4)*xinv(2,4)-+xinv(3,4)*xinv(3,4)-+xinv(4,4)*xinv(4,4); The finite element solution is store = node no. fem sol 1.0000 0.00000 2.0000 20.0000 3.0000 42.5000 4.0000 50.0000 5.0000 100.000 6.0000 42.50005.1 5.2 5.3 54 5.5 5.6 5.7 5.8 5.9 Problems 153 Problems Repeat Example 5.1.1 to derive Eq. (5.1.14). A square domain is modeled using either one bilinear element or two linear triangular elements as shown in Fig. P5.2. Compute the system matrix for the Laplace equation for each discretization. Flux through an element boundary is shown in Fig. P5.3. Determine the equivalent nodal fluxes. A uniformly distributed flux is given on a side of a biquadratic element as shown. in Fig P5.4. Compute the boundary integral to determine the equivalent nodal flux. The interpolation functions for the boundary nodes are F(e—1e-2) H(z) = 2(2- 2) A(z Ha(2) = hols — 1) A linear triangular element has three vertices like (#1, y1), (#2,y2) and (23, ys) Evaluate Eq. (5.2.26) for the element vector if a concentrated source of magnitude Q is located at (z,,y,) which lies within the element. Explain how to incorporate the boundary condition given at the edge of Fig. P5.6 into the finite element equation for the Laplace equation. Apply the Galerkin method and the Crank-Nicolson method to solve the following parabolic partial differential equation. du 1 du St ~ 0ae? = 10 0<2<3 Initially u is 50 all over the domain and the domain is subjected to boundary conditions u=100 at the left end and 24 = 100e~t at the right end. Using At=1, find the nodal solution at time t=1. The domain is discretized into two linear elements. As a result, the three nodal points are located at 71=0, z2=1, and z3=3, respectively. Redo Prob, 5.7 using the backward difference method. For a thermally orthotropic material, the two-dimensional heat conduction equation is a ou a (5 és) + By (Hy) += ° where ky and ky are heat conduction coefficients along the orthotropic axes 2 and y, respectively. Q is heat generation per unit volume. Develop the element matrix equation using linear triangular elements.154 Laplace's and Poisson’s Equations Chapter 5 Ww pry Figure P5.2 Problem 5.2 }-—_1__-.] domain Figure P5.3 Problem 5.3 ¥ @ = constant Figure P5.4 Problem 5.4 5.10 Repeat Example 5.9.1 for various mesh patterns shown in Fig. P5.10 using the computer program. Compare the salutions at the center of the domain. 5.11 A domain is normalized such that 0< z <1 and 0< y <1. Solve the LaplaceProblems 155. $e =-10U+Sxe-10 Figure P6.6 Problem 5.6 Figure PG.10 Probie 5.19 equation over the domaio using the provided program and 16 bilinear elements (4 elements in the x and y-axis, respectively), The boundary conditions are u=0 at 2=0, U=100 at e=1, u=D at yD, and u=200 at y=) 5.12 Solve the Laplace equation for the demain shown in Fig, P§.12. The boundary condition is 4lso shown in the figure. 5.13 Redo Prob. §.11 using the transient analysis assuming initia! condition u=0 all over the domain. Use the forward difference time integration technique 5.14 Redo Prob. 5,11 using the transient analysis assuming initial condition u=0 all over the domain. Use the backward difference time integration technique. §.15 Redo Prob. §.11 using the transient analysis assuming initis] condition u=0 all over the domain, Use the Crank-Nicolson time integration technique. 5.16 Redo Prob. 5.12 using the transient analysis with initial condition u=400. Use the Crank-Nicolson technique for time integration.186 Laplace's and Poisson's Equations Chapter 5 Insuoted on Figure PS.12 Problem 5.12 R20 eet I) am TnsvkatedCHAPTER SIX ISOPARAMETRIC ELEMENTS 6.1 One-Dimensional Elements [eoparametric elernents [10] use mathematical mapping from one coordinate system into the other coordinate system. The former coordinate system ia called the neturnl coordinate system while the lalter is called the physical coordinate system. The problem domain is provided in the physical coordinate system denoted xyz-axes in the following discussion. On the other hand, element shape functions are defined in terms of the natural coordinate system denoted €{-axee. As a result, mapping is needed between the two coordinate systems. ‘We consider a linear one-dimensional isoparametric element to discuss the basic characteristics of isoparametric elements. Multi-dimensional isoparametric elements will be discussed in the subsequent sections. Shape functions for the isopatametric element are given in terms of the natural coordinate system as seen in Fig. 6.1.1. The two nodes are located at £ = —1.0 and & = 1.6. These nodal positions are arbitrary but the proposed selection is very useful for numerical integration because the element in the natural coordinate system is normalized between -] and 1. The shape functions can be written aa tig) = 30-8) (BAM) and ' Ha{G) = 5(1 +8) (8.1.2) The physical linear element may be located at any position in the physical coordinate system aa shown in Fig. 6.1.2, The element has two nodal coordinate values x, and ‘2z with corresponding nodal variables u; and up. Any point between € = —1.0 and & = 1.0 in the natural coordinate system can be mapped onto a point between x; and zz in the physical coordinate system nsing the shape functions defined in Eqs (6.1.1) and (6.1.2). z= Hil€)2i + Ha(€)z2 (6.1.3) 17158 Iscparametric Elements Chapter 6 node ft node 2 t ——_+—__—__+—___- #=-t t=? Figure 6.1.1 Linear Element In the Natural Coordinate Figure 6.1.2 Linear Element in the Physical Coordinate The eame shape functions are also used to interpolate the variable u within the element us Hi(g)us + Halé)us (6.1.4) If the same shape functions are used for the geometric mapping as well at nodal variable interpolation such as Eqe (6.1.3) and (6.1.4), the element is called the isoparametric element. In order to compute #2, which is necessary in most cases to compute ¢lement masrices, we use the chain rule such that HO, , SHD,, = HOd,, + te) fu (6.1.5) where the expression requires $ which ia the inverse of ¥. The latter can be computed from Eq. (6.1.3). ad an, = ani, : oO), . flea a) 6.16) Substituting Eq. (6.1.6) into Ea, (6.1.5) yields ao tne ae. (6.1.7) Ae a result, derivatives of shape functions with reapect to the physical coordinate eyetem are ani dz mom ue) aie) de mm (6.1.8) in which A; = (22 — 21) ia the element size in the physical coordinate system. ‘These detivative values are identical to those obtained directly from the linear shape functions expressed in terms of the physical coordinate system like Eqs (2.4.7) and (2.4.8).Section 6.1 One-Dimensional Elements 159 Figure @.1.3 Quadratic Isoparametric Element Let us compute the following integral using the linear isoparametric element. fe (2S swu)ae (6.1.10) The integration is in terms of the physical coordinate system while the integrand is expressed in terms of the nafuraf coordinate aystem because isoparametric shape functions are used for the trial and test functions u and w. Hence, we want to write the integral in terms of the nadura/ coordinate system. To this end, we obtain [ (B8+6 1) 2a (6.1.11) where J = 4 is called the Jacobian. Substitution of the isoparametric shape functions for both u and w resulta in LGs[e te ter) )ee(e} Lk BVIG) wo ate hte This expression is the same as that obtained from the conventional linear element. At this point, the isoparametric clement does not seem to have an advantage over the conventional element because the isoparametric element requires more procedures such as mapping and chain rule. The major advantage of isoparametric elements comes when analytical integration to compute element matrices and column vectors is either very complicated or almost impossible. This is the case either element shapes in the physical dormain are oot regular such as in the rmulti-dimensional problem or the differential equation is quite complex. Therefore, (he numerical integration technique is needed. Because each isoparametric element is defined in terma of the normalized domain such a8 £ = —1 and £j = 1, it is much easier to apply any numerical integration technique. The application of numerical integration technique is discussed later in this chapter. @Example 6.1.1 Let us consider a quadratic one-dimensional isoparametric clement ag seen in Fig 6.1.3, Shape functions for this element are @-8 z Ai(E) = (6.1.13}160 Isoparametric Elements Chapter 6 #,(é)=1-@ (6.1.14) and Hs(€) = @+8 (6.1.15) ‘The variable u can be interpolated using these shape functions. u= Hy(€)us + Ho(é)u2 + Ha(é)us (6.1.16) Geometric mapping from the natural coordinate to the physical coordinate is 2 = Hy(€)e1 + Ho(é)22 + Ho(€)z9 (6.1.17) The Jacobian becomes ae dé o. dHi(€) J ae t; = (€-0.5)z — 2x2 + (+0.5)25 (6.1.18) i= If the mid-node 22 is located between the two end-nodes 2; and 23 (ie. 22 = ©1429)), the Jacobian becomes 4+ in which h; = £3 — 2; is the element length. Derivatives of the shape functions, Eqs (6.1.13) through (6.1.15), are ay) _ diy _ 1 ie TE = 5-0 (6.1.19) dHy(6) _ dH __4e ee (6.1.20) dH3(€)__1dH3 1 “al oa = B+ (6.1.21) 6.2 Quadrilateral Elements ‘The shape functions for the bilinear isoparametric element, are given below: 1 An(E,m) = g(1— (1-0) (6.2.1) th{Gn) = 70 +9000) (62.2) Hg(6,n) = F040 +0) (623)Section 6.2 Quadrilateral Elements 161 Figure 6.2.1 Bilinear Element in the Natural Coordinate (23,4) (anu, (22, Ue) (4%) Figure 6.2.2 Bilinear Element in the Physical Coordinate Han) = 40-900) (62.4) for the nodes shown in Fig. 6.2.1. These shape functions are defined in terms of the normalized natural domain (ie. ~1 < € <1 and -1< <1). While the element shape is a square in the natural coordinate system, it can be mapped into a general quadrilateral shape with distortion as seen in Fig. 6.2.2. When this mapping is undertaken, the relative positions of nodal points should be consistent between the two elements in the natural and physical domains. In other words, the second node is next to the first node in the counter-clockwise direction and similarly for the rest of the nodes. Then, a point (€,7) within the natural element is mapped into a point (z,y) within the physical element using the shape functions given in Eqs (6.2.1) through (6.2.4) as shown below: 4 =o Hien) (6.2.5) rf 4 y= Ain) (6.2.6) is162 Isoparametric Elements Chapter 6 in which 2; and y, are the coordinate values of the i** node. Similarly, any physical variable can be interpolated using the same shape functions. 4 u= SHE nu (6.2.7) in which u; is the nodal variable at node i. Let us apply this bilinear isoparametric element to the Laplace equation dis- cussed in Chapter 5. Then, we need to compute 2462) and 2462), respectively. In order to compute these derivatives, we use the chain rule, again. Oo _ bbe, bby BE = Be aE * By dE (628) 000s, aby 35 7 82 On * By On Rewriting these in the matrix form provides (2)-18 af=|ar 4 & } (6.2.10) on band bay Here, the derivative shown in the left-hand-side column vector is called local deriva- tive while that in the right-hand-side column vector is called global derivative. Fur- thermore, the square matrix in this equation is called the Jacobian matrix for the two-dimensional domain and denoted as (6.2.9) Jun Jn] [8% & Ae = 6.2.11 v=[3 e)=[8 & Sy The Jacobian matrix can be easily extended for the three-dimensional domain. Inverse of the Jacobian matrix is denoted by i =py = [2 ral Ra Rar ae Then, Eq. (6.2.10) can be rewritten as a 2. ve | _ [Ru re { aE } { a } ~ [Ra Ral le (6.2.13) As a result, the derivatives of shape functions with respect to x and y can be obtained from the above equation. BH; OH, ez \ _ [Ru Rar i} (H}= [Be Be] (239 IsSection 6.2 Quadrilateral Elements 163 (ad) (od) (a,b) (e.b) 2 Figure 6.2.3 Rectangular Element The components in the Jacobian matrix are computed as shown below: eo an one), . (6.2.15) i te x oH, 1) . (6.2.16) . Ray ony ED (6.2.17) ce be > $286, n) | (6.2.18) 1 Substitution of bilinear shape functions, Eqs (6.2.1) through (6.2.4), into the above expressions yields T= ia —n)ei+ ia = n)z2+ ia +n)z3— 5a +n)ea (6.2.19) Ja =—FO- mu + Z0= nant 0am FO + nue (62.20) Jn = Her H(t Gea + FL 4 Os + Hl - fee (6.2.21) Jn =-H0-On-F0FOm+ FIFO v4 TA-Ou (6.2.22) These components are in general a function of € and n. However, they may be constant for a special case as shown in the following example. Once the Jacobian matrix is computed from Eqs (6.2.19) through (6.2.22), global derivatives of shape functions are computed as as (6.2.28) outa a. BEG a + Ry) (6.2.24) OH) _ Ry He) + RD)164 Isoparametric Elements Chapter 6 & Example 6.2.1 Let us compute the Jacobian matrix for the physical element shown in Fig. 6.2.3. Substituting the nodal coordinate values into Eqs (6.2.19) through (6.2.22) yields the following matrix. = (F | (6.2.25) ‘As seen in this example, the Jacobian matrix becomes a diagonal matrix (i.e. all off-diagonal components vanish) when the element in the physical domain is a rectangular shape. In addition, the diagonal components are constant not a function of € and 7. The inverse of the Jacobian matrix becomes 3 (0 =| e-4 2. (2) [ 7 2 (6.2.26) The global derivatives of shape functions become (62.27) (6.2.28) oe = We (6.2.29) os = me (6.2.30) ee =-3a = (6.2.31) = Ha (6.2.32) ae = Cea (6.2.33) os 7 Wen (6.2.34) & Example 6.2.2 Let us compute the following integral using the same element as given in Example 6.2.1. Le y+ He) dedy (6.2.35)Section 6.2 Quadrilateral Elements 165 On. #) 3) 24 01,-%) Figure 6.2.4 Element of Trapezoidal Shape Substitution of Eqs (6.2.27) and (6.2.31) into Eq. (6.2.35) results in dope 1 : 1 F i i lz Sarl 9? + qa pa - 8" |dedy (6.2.36) ‘The lower and upper limits of the integrals can be changed using dedy = \J\dEdn (6.237) where |J| is the determinant of the Jacobian matrix and is equal to ==") for the present element. That is, |.J| is a constant value for a rectangular shape of physical element. Then, we obtain Aa ae. LL laetap 9+ gt 07] oe Dicin (6.2.88) Integration of Eq. (6.2.87) finally yields “2PHe-H" # Example 6.2.3. Fine the Jacobian matrix for the quadrilateral element shown in Fig. 6.2.4. Equations (6.2.19) through (6.2.22) along with the nodal coordinate values as specified in the figure yield 1 40+ 3| J] = [ 6.2.39) Vl=]o tis4e) ( ) The determinant of Jacobian is |J| = See which is always positive for —1<€ <1. Inverting the matrix gives (6.2.40)166 Isoparametric Elements Chapter 6 Figure 6.2.5 Element of Quadrilateral Shape ‘The determinant of the Jacobian matrix is |J| = }(3 + €). In order to compute the integration as given in Eq. (6.2.35) for the present element, we first compute arn Ox (6.2.41) (6.2.42) ‘The expression for the integral becomes ie [f-pge Cote (E11) St Dace (6.2.43) ‘This integral can be conducted analytically. However, if the shape of the physical element has more severe distortion, the integral becomes more complicated and may be beyond the analytical computation. Even if analytical integration may be possible, performing the analytical integration for every element of different shape is not pratically possible. Therefore, the numerical integration technique is used along with the isoparametric element. & Example 6.2.4 — The physical element has a severe distortion as seen in Fig. 6.2.5. The corresponding Jacobian matrix is 1-32 | (6.2.44) 7 wi=[4, Se 14 and its determinant is 3(2—3€ —3n). This determinant can be zero or negative for -1 < € < 1 and —1 < 9 < 1. Hence, this shape of element should be avoided in discretizing the physical domain. +Section 6.2 Quadrilateral Elements 167 Some other popular quadrilateral isoparametric elements are eight-node and nine-node elements as shown in Fig. 6.2.6. Their shape functions are given below. Eight-node element: Nine-node element: = }0-90-n(-1-€-0) Hy = H+ 90-1 +E-) Hy = Fa+9+n\(-1+e+m) y= fa-9a +n)(-1-£+9) Hy = h0-@)1-1) He= hasan) a= }a-e)a4n) we=ia-9a-1) he ie - 6x? ~ 0) Hy = HE +O(0?-1) Hy = E4607 +1) Hy= Hey" +1) Hy = 30 -@)(7? =) He= HE +002) Hy =$- E07? +9) We=1@-90-17) Hg = (1-€*)(1—17) (6.2.45) (6.2.46) (6.2.47) (6.2.48) (6.2.49) (6.2.50) (6.2.51) (6.2.52) (6.2.53) (6.2.54) (6.2.55) (6.2.56) (6.2.57) (6.2.58) (6.2.59) (6.2.60) (6.2.61)168 Isoparametric Elements Chapter 6 n Figure 6.2.6 Eight-Node Isoparametric Element Figure 6.2.7 Nine-Node Isoparametric Element 6.3 Triangular Elements Like quadrilateral isoparametric elements, triagular isoparametric elements can be defined. Shape functions of the linear triangular element are in terms of the natural coordinate system My =1-€-n (6.3.1) Hy =€ (6.3.2) Hs=n (6.3.3) for the nodes shown in Fig. 6.3.1. The quadratic triangular element has the following shape functions with reference to Fig. 6.3.2. Hy = (1—€ - )(1 — 2€ — 2n) (6.3.4) Hy = &(2€ - 1) (6.3.5) Hs = (2-1) (6.3.6) Hy = 4€(1—€—7) (6.3.7)Section 6.3 ‘Triangular Elements 169 2 f=1 Figure 6.3.1 Three-Node Triangular Element in the Natural Coordinate (0,0) (0.5,0) (1,0) Figure 6.3.2 Six-Node Triangular Element in the Natural Coordinate Hs = 4€q (6.3.8) He = 4n(1—~n) (6.3.9) & Example 6.3.1 Consider an element as shown in Fig. 6.3.3. The Jacobian matrix for the element is = f2-t wW-H m=[278 nae (6.3.10) and its determinant is |J| = (22 — 21)(ys — y1) — (@3 — 21)(yo — y1) which equals twice the triangular area in the physical domain. {170 Isoparametric Elements Chapter 6 Figure 6.3.3 Three-Node Triangular Element in the Physical Coordinate 6.4 Gauss Quadrature Integral is defined as 5 a f[ S(2)dz = im YY f(zi)das (6.4.1) ie i= This is shown in Fig. 6.4.1. In the numerical integration, we take a finite number of calculations. Therefore, Eq. (6.4.1) is approximated as b N f[ F(a)de = So flei)Azi (6.4.2) 7 i= where N is a finite number. Rewriting this expression in a general way gives 6 M [ F(z)de = S* f(2:) Wi (6.4.3) 7 i=l in which M is the number of integration points, 2; is the integration point (or sampling point), and W; is called the weighting coefficient. The weighting coefficient can be interpreted as the width of the rectangular strip whose height is f(z;) by comparing Eqs (6.4.2) and (6.4.3). Any numerical integration may be expressed in this form. In order to derive standard values for the integration points and weighting coefficients, the integration domain is normalized such that —1 < x <1. Of course, there are other ways for normalization. #& Example 6.4.1 Let us find the proper integration points and weighting coefficients for the two point trapezoidal rule. The trapezoidal rule gives [ : 9()dé = (g(—1) + 9(1)) (6.4.4)Section 6.4 Gauss Quadrature 171 Figure 6.4.1 Integration Comparing Eq. (6.4.4) to Eq. (6.4.3) indicates that the integration points for this case are 21 = —1 and zz = 1 while the weighting coefficients are W; = 1 and W,=1. ¢ 4 Example 6.4.2 Repeat Example 6.4.1 using Simpson’s 4 rule with three point integration. This integration results in ; J st6de = Glo(-2) + 49(0) +900) (6.4.5) Therefore, we obtain 2} = —1, 22 = 0, 23 = 1, Wi = 3, We = 4, and Gauss-Legendre quadrature is very useful for integration of polynomial func- tions. It can integrate a polynomial function of order 2n — 1 using the n-point quadrature exactly. Integration points and weighting coefficients for Gauss-Legendre quadrature are provided in Table 6.4.1. Similarly, Table 6.4.2 gives integration points and weighting coefficients for the triangular domain shown in Fig. 6.3.1 and Fig. 6.3.2. If the integrand is not a polynomial expression, Gauss-Legendre quadrature gives an approximate result. In this case, an optimal number of integration points should be selected in consideration of accuracy and computational cost. The next example shows how to determine the integration points and weighting coefficients for the Gauss-Legendre quadrature. 4 Example 6.4.3 This example shows a way how to compute the sampling points and weighting coefficients for Gauss-Legendre quadrature. Let us integrate172 Isoparametric Elements Chapter 6 Table 6.4.1 Sampling points and weights in Gauss-Legendre numerical integration 2 Int.Point Weight 1 0. 0.00000 00000 00000 2.00000 00000 00000 2 = +£0.57735 02691 89626 1.00000 00000 00000 3 +£0.77459 66692 41483 0.55555 55555 55556 0.00000 00000 00000 0.88888 88888 88889 4 +0.86113 63115 94053 0.34785 48451 37454 0.33998 10435 84856 0.65214 51548 62546 5 0.90617 98459 38664 0.23692 68850 56189 +£0.53846 93101 05683 0.47862 86704 99366 0.0000 00000 00000 0.56888 88888 88889 6 +£0.93246 95142 03152 0.17132 44923 79170 +0.66120 93864 66265 0.36076 15730 48139 0.23861 91860 83197 0.46791 39345 72691 a cubic polynomial as shown in Fig. 6.4.2. In Gauss-Legendre quadrature, we want to make the integration of the cubic polynomial the same as that of a linear function. In other words, the two different hatched areas in Fig. 6.4.2 are the same (i.e. Area(A)=Area(B) in Fig. 6.4.2). Then we can write ie i: 2 [feds [aed = I) (6.4.6) where F(x) = a9 + a2 + agz? + agz® (6.4.7) 9(2) = co + et (6.4.8) and W, and 2, are the weighting coefficient and sampling point for the two point Gauss-Legendre quadrature because the two point rule integrates a cubic polynomial exactly. Let us rewite the cubic polynomial in the following way. F(z) = co +12 + (2 — 2)(z — 22)(bo + 612) (6.4.9) In this expression, 21 and 2» are fixed constants to be determined later. However, there are still four general constants co, c1, bo and 6; to be determined to makeSection 6.4 Gauss Quadrature 173 Table 6.4.2 Numerical integrations over triangular domains Int. order___€-coordinate n-coordinate Weight 0.16666 66666 667 0.16666 66666 667 0.33333 33333 333 S-points 0.66666 66666 667 0.16666 66666 667 0.33333 33333 333 0.16666 66666 667 0.66666 66666 667 0.33333 33333 333 0.10128 65073 235 0.10128 65073 235 0.12593 91805 448 0.79742 69853 531 0.10128 65073 235 0.12593 91805 448 0.10128 65073 235 0.79742 69853 531 0.12503 91805 448 T-points 0.47014 20641 051 0.05971 58717 898 0.13239 41527 885 0.47014 20641 051 0.47014 20641 051 0.18239 41528 885 0.05971 58717 898 0.47014 20641 051 0.13239 41528 885 0.33333 33333 333 0.33333 33333 333 0.22500 00000 000 0.06513 01029 022 0.06513 01029 022 0.05334 72356 088 0.86973 97941 956 0.06513 01029 022 0.05334 72356 088 0.06513 01029 022 0.86973 97941 956 0.05334 72356 088 0.31286 54960 049 0.04869 03154 253 0.07711 37608 903 0.63844 41885 698. 0.31286 54960 049 0.07711 37608 903 13-points 0.04869 03154 253 0.63844 41885 698 0.07711 37608 903 0.63844 41885 698. 0.04869 03154 253 0.0711 37608 903 0.31286 54960 049 0.63844 41885 698 0.0711 37608 903 0.04869 03154 253. 0.04869 03154 253 0.0711 37608 903 0.26034 59660 790 0.26034 59660 790 0.17561 52574 332 0.47930 80678 419 0.26034 59660 790 0.17561 52574 332 0.26034 59660 790 0.47930 80678 419 0.17561 52574 332 0.33333 33333 333 0.88333 33833 333. -0.14957 00444 67 Eq. (6.4.9) the same as Eq. (6.4.7) for arbitrary constants a;. Substitution of Eq. (6.4.9) into Eq. (6.4.6) states [ *(e—21)(2— 22)(b0 + bi2)d2 =0 (6.4.10) :174 Isoparametric Elements Chapter 6 B Gf )=00+ ey Sa)=ap+ a2 + a22?+ age* A Pte “=f 1 Figure 6.4.2. Two-Point Gauss-Legendre Quadrature Equation (6.4.10) must be true independent of bo and 6; because the integration rule holds for a general cubic polynomial. Therefore, 4 iy (2 —21)(2—22)d2 = 0 (6.4.11) =1 and 1 f{ 2(0 — 21)(e — 22)de = 0 (6.4.12) aa ‘These two equations determine 2) = —Js and 22 = a ‘These are two sampling points for the two-pont Gauss-Legendre quadrature. In order to find the corresponding weighting coefficients, we integrate 1 I | (co + c12)dz = 2c (6.4.13) 1 From Eq. (6.4.6), this integration is equal to 2 T= DWF (22) = Walco + crzr) + Walco + e122) i = ¢o(Wi + Wa) — Fa —We) (6.4.14) Equating Eq. (6.4.13) and Eq, (6.4.14) yields two weighting coefficients W, = 1 and W=1. 4 Example 6.4.4 Perform the following integral: [cites setae (6.4.18)Section 6.4 Gauss Quadrature 175 Because the order of polynomial is 2, 2n — 1 = 2. From this, we get n = 1.5 ‘The number of integration points should be an integer. So, we use the two point quadrature rule. From Table 6.4.1, the two integration points are “KR and Jz with weighting coefficient 1 for each point, respectively. The numerical integration becomes ‘ iL (1+ 2€ + 3€%)de = o (I+ ARF) +I T) + LAI +AGI=A 6A.16) ‘This is the exact solution. If we use the three point quadrature to integrate Eq. (6.4.15) (ie. & = —“48, & = 0, & = GS, W = §, We = 8, and Ws = 8), we also obtain the same exact solution. Therefore, the quadrature rule using two or higher number of integration points will always yield the exact solution for this problem. + The quadrature rule can be extended for multi-dimensional integration. For ex- ample, numerical integral in the normalized two-dimensional domain can be conducted in the following way. ce ff senedn fe 1M. =/ DY Wial&sn)dn 7 Ma M = Ly DY Wis(€75) fee Mi Ma = VL wWj9(G 15) (6.4.17) 421551 in which M, and M2 are the number of integration points in the € and 7 axes, respectively. In addition, (&;,n;) are the integration points and W; and W; are weighting coefficients. Table 6.4.1 can be used for these values. Similarly, numerical integration in three-dimension becomes 1 1 us Mi Mz Ms i I, i a 96, ,C)ddnde = S11 S| Wis Wea Gis m5) (6.4.18) j=1k=1 4 Example 6.4.5 Integrate the following expression: [ ; f ; 9€ paddy (6.4.19)176 Isoparametric Elements Chapter 6 ‘The integrand is the second order in terms of € and 1, respectively. That is, 2M, —1 = 2Mz—1 = 2 for both axes. Therefore, we use two point quadrature in both € and 7 directions. The integration points are £; = m1 = -# and & = jm = Te The weighting coefficients are W,; = W, = 1 and Wo = Wo = 1. Applying these values to Eq. (6.4.17) results in 4. t & Example 6.4.6 We want to integrate [ : [ : 15€2n*dédy (6.4.20) ‘The integrand is the second order in terms of € and fourth order in terms of n. Therefore M; = 2 and Mz = 3. Using the two point quadrature in the € direction and three point quadrature in the 7) direction from Table 6.4.1, we obtain the solution of 4. { 6.5 MATLAB Application to Gauss Quadrature This section shows MATLAB examples for numerical integration of one-, two- or three-dimensional functions using Gauss-Legendre quadrature. The domain of integration is normalized between -1 and 1 for every axis. # Example 6.5.1 We want to integrate F(x) = 142? — 329 4425 (6.5.1) over the domain —1 < z < 1 using Gauss-Legéndre quadrature. Because the highest order of polynomial is 5, we need the 3-point quadrature rule for exact integration from 2n — 1 = 5. The numerical result is §. The MATLAB program is shown below. %e % Example 6.5.1 % Gauss-Legendre quadrature of a function in 1-dimension % % Problem description % Integrate f(x)=1+x?-3x344x* between x=-1 and x=1 % Variable descriptions % point1 = integration (or sampling) points % weight! = weighting coefficients % ngl = number of integration pointsSection 6.5 MATLAB Application to Gauss Quadrature 177 %- % ngl=3; % (2*ngl-1)=5 % [point ,weight1]=feglqd1 (ngl); % extract sampling points and weights % % % summation for numerical integr: % % value=0.0; % for int=1:1 x=point (int); wt=weight1 (int); +x3-3*x3444x5; % evaluate function at sampling point ‘value+func*wt; % print the solution function [point1,weight1]=feglqdi (ngl) %- % Purpose: % determine the integration points and weighting coefficients % of Gauss-Legendre quadrature for one-dimensional integration % % Synopsi % (point ,weight1]=feglqd1 (ngl) h % Variable Description: % ngl - number of integration points % point - vector containing integration points % weight! - vector containing weighting coefficients ee wh % initialization % point1=zeros(ngl,1); weight1=zeros(ngl,1); % % find corresponding integration points and weights % if ngl=: % 1-point quadrature rule178 Isoparametric Elements Chapter 6 pointl(1)=0.0; weight}(1)=2.0; % elseif ngl==2 % 2-point quadrature rule point1(1)=-0.577350269189626; point (2)=-point1(1); weight1(1)= weight1(2) % 3-point quadrature rule 0.774596669241483; point1(3)=-point1(1); weight 1 (1)=0.555555555555556; weight (2)=0.888888888888889; weightl(3)=weight1(1); % elseif ngl==4 % 4-point quadrature rule point1(1)=-0.861136311594053; point (2)=-0.339981043584856; point] (3)=-point1(2); point1(4)=-pointl(1); weight1(1)=0.347854845137454; weight1(2)=0.652145154862546; weight] (3)=weight1 (2); weight1(4)=weight1(1); % else % 5-point quadrature rule point] (1)=-0.906179845938664; point] (2)=-0.538469310105683; point1(3)=0.0; point] (4)=-point1(2); point1(5)=-pointl (1); weight (1)=0.236926885056189; weight (2)=0.47862867049936 weight 1(3)=0.568888888888889; weight1(4)=weight1(2); weight1(5)=weight1(1); % end %—Section 6.5 MATLAB Application to Gauss Quadrature 179 & Example 6.5.2 Use Gauss-Legendre quadrature for integration of J(z,y) = 1+ dey — 327y? + 2ty® (6.5.2) over the domain —1 < z < 1 and —1 < y < 1. We use 3.point quadrature rule along the x-axis and 4-point quadrature rule along the y-axis. The result is 2.7810. ~%— % Example 6.5.2 % Gauss-Legendre quadrature of a function in 2-dimension % % Problem description % Integrate {(x,y)=1+4xy-3x?y?-+x4y® over -1 ngly ngl=nglx; else ngl=ngly; end % % initialization % point2=zeros(ngl,2); weight2=zeros(ngl,2); h % find corresponding integration points and weights % {pointx,weightx]=feglqd1 (nglx); [pointy,weighty]=feglad1 (ngly); % % quadrature for two-dimension % for intx=1:nglx point2(intx,1)=pointx(intx); weight2(intx,1)=weightx(intx); end % for inty=I:ngly point2(inty,2)=pointy(inty); weight2(inty,2)=weighty (inty); end % %Section 6.5 MATLAB Application to Gauss Quadrature 181 4 Example 6.5.3 The following three-dimensional function is integrated using Gauss-Legendre quadrature. F(2,y,z) = 1+ 4n?y? — 30224 + yte® (6.6.3) over the normalized domain —1 < z < 1, —1 < y< land -1 ngly if nglx > nglz ngl=nglx; else ngl=nglz; end else if ngly > nglz ngl=ngly; else ngl=nglz; end end % ‘% initialization % point3=zeros(ngl,3); weight3=zeros(ngl,3); % % find corresponding integration points and weights %Section 6.6 MATLAB Application to Laplace Equation [pointx, weightx]=feglad1 (nglx); [pointy,weighty]=feglqd1(ngly); % quadrature rule for x-axis % quadrature rule for y-axis 183 [pointz, weightz]=feglqd1(nglz); % quadrature rule for z-axis % % quadrature for two-dimension % quadrature in x-axis pointx(intx); =weightx(intx); end for inty=1: % quadrature in y-axis point3(inty,2)=pointy(inty); weight3(inty,2)=weighty (inty); end % for intz=1:nglz % quadrature in z-axis point3(intz,3)=pointz(intz); weight3(intz,3)=weightz(intz); end % %— 6.6 MATLAB Application to Laplace Equation Isoparametric elements are used to solve the Laplace equation which was discussed in Chapter 5. #& Example 6.6.1 This example shows how to compute the element matrix for the Laplace equation. The clement matrix is expressed as Ree cael Ine | Oz Ox dy dy ‘The element domain is shown in Fig. 6.2.4. The MATLAB program is listed below to evaluate the element matrix. (6.6.1) %— % Example 6.6.1 % Compute element matrix for two-dimensional Laplace equation % % Problem description % Determine the element matrix for Laplace equation using % isoparametric four-node quadrilateral element and Gauss-Legendre184 Isoparametric Elements % quadrature for a single element shown in Fig. 6.2.4. % % Variable descriptions % k - element matrix % point2 - integration (or sampling) points % weight2 - weighting coefficients % nglx - number of integration points along x-axis % ngly - number of integration points along y-axis % xcoord - x coordinate values of nodes % ycoord - y coordinate values of nodes % jacob2 - Jacobian matrix % shape - four-node quadrilateral shape functions Chapter 6 % dhdr - derivatives of shape functions w.r.t. natural coord. r % dhds - derivatives of shape functions w.r. natural coord. s % dhdx - derivatives of shape functions w.r.t. physical coord. x % dhdy - derivatives of shape functions w.r.t. physical coord. y %- % nnel: % number of nodes per element, ndof=1; % degrees of freedom per node edof=nnel*ndof; % degrees of freedom per element % nglx=2; ngl % use 2x2 integration rule % xcoord=[-1 1 1-1}; % x coordinate values ycoord=[-0.75 -0.75 1.25 0.25); % y coordinate values % [point2,weight2]=feglqd2(nglx,ngly); % sampling points & weights % os % numerical integra %— k=zeros(edof,edof); % initialization to zero % for intx=I:nglx x=point2(intx,1); % sampling point in x-axis wix=weight2(intx,1); % weight in x-axis for inty=1:ngly y=point2(inty,2); % sampling point in y-axis wly=weight2(inty,2) ; % weight in y-axis % [shape,dhdr,dhds]=feisoq4(x,y); % compute shape functions and % derivatives at sampling point % jacob2=fejacob2(nnel,dhdr,dhds,xcoord,ycoord); % compute Jacobian detjacob=det(jacob2); % determinant of Jacobian % inverse of Jacobian matrixSection 6.6 MATLAB Application to Laplace Equation 185 {dhdx,dhdy]=federiv2(nnel,dhdr,dhds,invjacob); % derivatives w.r.t. % physical coordinate % %— % element matrix loop % for i=Liedof for j=1:edof k(i,j)=k(i,j) + (dhdx(i)*dhdx(j)+dhdy (i)*dhdy(j))*wtx*wty*detjacob; end end % end of numerical integration loop % print the element matrix RRO wWG Sx are function [dhdx,dhdy]=federiv2(nnel,dhdr,dhds,invjacob) %: % Purpose: % determine derivatives of 2-D isoparametric shape functions with % respect to physical coordinate system % % Synopsis: % (dhdx,dhdy)=federiv2(nnel,dhdr,dhds,invjacob) % % Variable Description: % dhdx - derivative of shape function w.r. | coordinate x % ahdy - derivative of shape function w.r. % nnel - number of nodes per element % dhdr - derivative of shape functions w.r.t. natural coordinate r % dhds - derivative of shape functions w.r.t. natural coordinate s . physi . physical coordinate y % invjacob - inverse of 2-D Jacobian matrix dhdy (i)=invjacob(2,1)*dhdr(i)+invjacob(2,2)*dhds(i); end nh %186 Isoparametric Elements Chapter 6 function [shapeq4,dhdrq4,dhdsq4]=feisoq4(rvalue,svalue) % % Purpose: % compute isoparametric four-node quadrilateral shape functions % and their derivatves at the selected (integration) point % in terms of the natural coordinate % % Synopsis: % [shapes dhdrq4,dhdsq4]=feisoq4(rvalue,svalue) % % Variable Description: % shapeqd - shape functions for four-node element % dhdra4 - derivatives of the shape functions w.r.t. r % dhdsqé - derivatives of the shape functions w.r.t. 8 % rvalue - r coordinate value of the selected point % svalue - s coordinate value of the selected point % % Notes: % Ast node at (-1,-1), 2nd node at (1,1) % 3rd node at (1,1), 4th node at (-1,1) % % % shape functions % shapeq4(1)=0.25*(1-rvalue)*(1-svalue); shapeq4(2)=0.25*(1+rvalue)*(1-svalue); shapeq4(3)=0.25*(1-+rvalue)*(1-+svalue); shapeq4(4)=0.25*(1-rvalue)*(1+svalue); % % derivatives % dhdrq4(1)=-0.25*(1-svalue); dhdrq4(2)=0.25*(1-svalue); dhdrq4(3)=0.25*(14svalue); dhdrq4(4)=-0.25*(1+svalue); % dhdsq4(1)=-0.25*(1-rvalue); .25*(14rvalue); dhdsq4(3)=0.25¢(1+rvalue); dhdsq4(4)=0.25*(1-rvalue); % %— % Purpose:Section 6.6 MATLAB Application to Laplace Equation. 187 % determine the Jacobian for two-dimensional mapping % % Synopsis: % [jacob2]=fejacob2(nnel,dhdr,dhds,xcoord,ycoord) % % Variable Description: % jacob2 - Jacobian for one-dimension % nnel - number of nodes per element % dhdr - derivative of shape functions w.r.t. natural coordinate r % dhds - derivative of shape functions w.r.t. natural coordinate s % xcoord - x axis coordinate values of nodes % ycoord - y axis coordinate values of nodes %— % jacob2=zeros(2,2); jacob2(1,1)+dhdr(i)*xcoord(i); jacob2(1,2)+dhdr(i)*ycoord(i); jacob2(2,1)+dhds(i)*xcoord(i); The computed element matrix is 0.7500 0.0000 + —0.2500 —0.5000 0.0000 0.7500 + —0.2500 —0.5000 —0.2500 —0.2500 0.5000 0.0000 —0.5000 -—0.5000 0.0000 1.0000 [K]= (6.6.2) & Example 6.6.2 Repeat Example 5.9.2 using isoparametric elements. Four-node quadrilateral elements are used. The finite element solution is the same as that obtained in Example 5.9.2. As a result, the solution is not repeated here. The MATLAB program is shown below. % Example 6.6.2 % to solve the two-dimensional Laplace’s equation given as % w,xx + wyy =0,0 node no. and j-> x or y %Section 6.6 MATLAB Application to Laplace Equation 189 gcoord(7,1)=1.25; gcoord(7,2)=2.5; gcoord(8,1)=2.5; gcoord(8,2)=2.5; gcoord(9,1 gcoord(10,1)=5.0; gcoord(10,2)=2.5; gcoord(11,1)=0.0; gcoord(11,2)=5.0; gcoord(12,1)=1.25; gcoord(12,2)=5.0; gcoord(13,1)=2.5; gcoord(13,2)=8.0; gcoord(14,1)=3.75; gcoord(14,2) gcoord(15,1)=5.0; gcoord(15,2)=5.0; gcoord(16,1)=0.0; gcoord(16,2)=7.5; geoord(17,1)=1.25; gcoord(17,2)=7.5; gcoord(18,1)=2.5; gcoord(18,2)=7. geoord(19,1)=3.75; gcoord(19,2 gcoord(20,1)=8.0; gcoord(20,2)=7. gcoord(21,1)=0.0; gcoord(21,2)=I10. gcoord(22,1)=1.25; gcoord(22,2 gooord(23,1)=2.5; gcoord(23,2)=10. gcoord(24,1)=3.75; gcoord(24,2)=10.; gcoord(25,1)=5.0; gcoord(25,2)=10.; % cece % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes %¥ nodes(1,1)=1; nodes(1,2)=2; nodes(1,3)=7; nodes(1,4)=6; nodes(2,1)=2; nodes(2,2)=3; nodes(2,3)=8; nodes(2,4): nodes(3,1)=3; nodes(3,2)=4; nodes(3,3)=9; nodes(3,4)=8; nodes(4,1)=4; nodes(4,2)=5; nodes(4,3)=10; nodes(4,4)=9; nodes(5,1)=6; nodes(5,2)=7; nodes(5,3)=12; nodes(5,4) nodes(6,1)=7; nodes(6,2)=8; nodes(6,3)=13; nodes(6,4) nodes(8,1)=9; nodes(8,2)=10; nodes(8,3)=15; nodes(8,4)=! nodes(9,1)=11; nodes(9,2)=12; nodes(9,3) nodes(10,1) nodes(11,1 19; nodes(11,4)=18; nodes(12,1 0; nodes( 12,4} nodes(13,1 22; nodes(13,4)=21; nodes(14,1 3; nodes(14,4)=22; nodes(15,1)=18; nodes(15,2)=19; nodes(15,3)=24; nodes(15,4)=23; nodes(16,1)=19; nodes(16,2)=20; nodes(16,3)=25; nodes(16,4)=24; % EE % input data for boundary conditions (ee % bedof(1)=1; % first node is constrained % whose described value is 0 % second node is constrained190 Isoparametric Elements Chapter 6 beval(2) % whose described value is 0 bedof(3)=: % third node is constrained % whose described value is 0 % 4th node is constrained ‘% whose described value is 0 % 5th node is constrained % whose described value is 0 bedof(6)=6; % 6th node is constrained % whose described value is 0 % 11th node is constrained beval(7)=0; % whose described value is 0 bedof(8)=16; % 16th node is constrained beval(8)=0; % whose described value is 0 bedof(9)=21; % 21st node is constrained % whose described value is 0 % 22nd node is constrained % whose described value is 38.2683 % 2rd node is constrained % whose described value is 70.7107 % 24th node is constrained % whose described value is 92.3880 % 25th node is constrained beval(13)=100; % whose described value is 100 % ———————————— % initialization of matrices and vectors beval(9): a ff=zeros(sdof,1); % initialization of system force vector kk=zeros(sdof,sdof); % initialization of system matrix index=zeros(nnel*ndof,1); % initialization of index vector % > —$— $$ % \oop for computation and assembly of element matrices %— ee [point2,weight2]=feglqd2(nglx,ngly); % sampling points & weights % for iel=1:nel % loop for the total number of elements % for i=1:nnel nd(i)=nodes(iel,i); % extract connected node for (iel)-th element xcoord(i)=gcoord(nd(i),1); % extract x value of the node ycoord(i)=gcoord(nd(i),2); % extract y value of the node end % k=zeros(edof,edof); ialization of element matrix to zero % %— % numerical integrationSection 6.6 MATLAB Application to Laplace Equation 191 %——_—. for intx=1:nglx x=point2(intx,1); % sampling point in x-axis wtx=weight2(intx,1); % weight in x-axis for inty=l:ngly y=point2(inty,2); % sampling point in y-axis wty=weight2(inty,2) ; % weight in y-axis % [shape,dhdr,dhds]=feisoq4(x,y); % compute shape functions and % derivatives at sampling point % jacob2=fejacob2(nnel,dhdr,dhds,xcoord,ycoord); % compute Jacobian % detjacob=det (jacob2); % determinant of Jacobian invjacob=inv(jacob2); % inverse of Jacobian matrix % [dhdx,dhdy]=federiv2(nnel,dhdr,dhds,invjacob); % derivatives w.r.t. % physical coordinate % —— % compute element matrix k(i,j)=k(ij) + (dhdx (i)*dhdx(j) +dhdy (i)*dhdy(j))*wtx*wty*detjacob; end end % end end % end of numerical integration loop % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % %— a % assemble element matrices ————_— % kk=feasmblI (kk,k,index); % end % %— % apply boundary conditions %—__—_—____— [kk, ff] =feaplyc2 (Kk, ff, bcdof, beval); % G————$_—_—_—___—_— % solve the matrix equation %—192 Isoparametric Elements fsol=Kk\ ft; % ae % analytical solution % for i=I:node x=gcoord(i,1); y=gcoord (1,2); esol(i)=100*sinh(0.31415927*y)*sin(0.31415927*x) /sinh(3.1415927); end % G3 —_- % print both exact and fem solutions Chapter 66.1 6.2 6.3 6.4 6.5 6.6 6.7 68 Problems 193 Problems Compute the following integral using the quadratic isoparametric element: b= [BY sw The shape functions are given in Eqs (6.1.13) through (6.1.15) and the element has nodes z;=2, z2=4, and 23=6 in the physical coordinate system, Consider one-dimensional isoparametric shape functions as given in Eqs (6.1.13) through (6.1.15). The isoparametric element is mapped into a physical domain with nodal points located at 2; = 0, z2 =a, and zy = 4, where a = 1.5, a= 1, or a= 0.5. Compute the Jacobian J and its inverse for these cases. Compute the Jacobian matrix for the following bilinear element shown in Fig. P6.3 and evaluate Oth Oy , Oth 2) wan [Bee de * ay Oy) Using the isoparametric element and 3 by 3 Gauss-Legendre quadrature. The shape functions are provided in Eqs (6.2.1) through (6.2.4). For the linear triangular isoparametric element shown in Fig. P6.4, (a) compute the Jacobian matrix and (b) find 9 in which H; is given in Eq. (6.3.1). Evaluate the Jacobian matrix for the four-node element shown in Fig. P6.5 using the bilinear isoparametric element. Gauss-Legendre quadrature rule is used to evaluate the integral | ff e@naste miata in which H, and Hz are quadratic shape functions of € and 9, respectively. What order of integration is necessary for exact integration of the integral if the element has no distortion (i.e. a rectangular shape of element in the physical domain)? ‘Two different isoparametric elements are used together as shown in Fig. P6.7. There is an interelement boundary between (z,y)=(1,1) and (z, y)=(2,2). Show that variable is continuous across the interface boundary. In other words, variable interpolation from the quadrilateral element is the same as that from the triangular element at the interface. Consider two elements shown in Fig. P6.7 again. For the elements, we use the following interpolation for each element. For the triangular element, we use = ao +aiz tay194 Isoparametric Elements Figure P6.3 Problem 6.3 eA) (42) a Figure P6.4 Problem 6.4 and for the quadrilateral element we use usb +bzt boy tbsry Is u compatible at the element interface of the two elements? Chapter 6 6.9 Two kinds of quadrilateral isoparametric elements are utilized together to mesh a domain as seen in Fig. P6.9. One is a bilinear element and the other is a biquadratic element. Is it compatible between the element interface? 6.10 Solve Prob. 5.11 using isoparametric elements and computer programs provided in this chapter. 6.11 Solve Prob. 5.12 using isoparametric elements and MATLAB programs.Figure P6.5 Figure P6.7 Problems _ 45,2) (25,2) a e,) Ls Problem 6.5 << (2,2) ay x Problem 6.7 Figure P6.9 Problem 6.9 195CHAPTER SEVEN TRUSS STRUCTURES 7.1 One-Dimensional Truss ‘The one-dimensional truss is also called a rod or an axial bar which was described in Chapter 4. The governing equation to describe the motion of rod is derived below. Let E, A and p indicate the elastic modulus, cross-sectional area and density of the rod, respectively. Applying Newton’s second law to the free body diagram shown in Fig. 7.1.1 gives au ap pAds ay = (e+ a) -P (7.1.1) where u is the axial displacement along the rod direction, and z and ¢ are the spatial and temporal axes, respectively. Hooke’s law states P qa (7.1.2) The strain-displacement relation is Ou cae (7.1.3) Substituting Eq. (7.1.3) into Eq. (7.1.2) and the result into (7.1.1) yields ae Ga a ou pAaea = £ (s0%) (7.1.4) In Eq. (7.1.4), p, A and E may vary as a function of z. ‘The weak formulation for Eq. (7.1.4) is - eu Ow du dul” I =[ (raw? + ABS ae - [sewgel (7.1.5) 197198 ‘Truss Structures Chapter 7 Figure 7.1.1 Free Body Diagram for Axial Member in which w is the test function. The first term is the inertia term and the second term is the stiffness term. Discretization of the domain into a number of elements breaks the global integral in Eq. (7.1.5) into element integrals over the element domains. Use of Galerkin’s method and linear shape functions for a rod element whose length is I results in the following stiffness matrix: 1 aH, wel= [ae {ge} Mae (73.6) b for the element nodal degrees of freedom {u; ug} as shown in Fig. 7.1.2. Superscript e denotes element. Here H; is the linear shape function which is (7.1.2) Hy = ; (7.1.8) Substitution of these shape functions into Eq. (7.1.6) results in the element stiffness matrix for rod. [x] AE[1 -1 mi 4 l (7.1.9) This is the same as given in Chapter 4. ‘The element mass matrix for rod is obtained from the first term in Eq. (7.1.5) using the linear shape functions. [Me] oe [; al (7.1.10) for the constant density and cross-sectional area. This is called the consistent mass matrix for rod. The lumped mass matrix is pal (0 ‘| (7.1.11) ¢ MI="F lo 1 This is obtained by lumping the distributed mass within the element into the concentrated masses at the two nodal points as seen in Fig. 7.1.3.Section 7.2 Plane Truss 199 Figure 7.1.2 Two-Node Axial Bar PA LLLILLLILLLLLLLL PAL PAL S 4 Figure 7.1.3 Equivalent Spring-Mass System 7.2 Plane Truss ‘The truss is a structure which consists of axial members connected by pin joints. Therefore, each member of the truss structure supports the external load through its axial force and it does not undergo the bending deformation. The stiffness matrix for a truss member shown in Fig. 7.2.1 is given in Eq. (7.1.9). However, the matrix size becomes 4x 4 because the nodal degrees of freedom of the truss element are expressed as {d°}={ur v1 uz v2)" (7.2.1) Here superscript e denotes the element level. The corresponding stiffness matrix is k 0 -k 0 0000 [l=], 9 bo (7.2.2) 0000 in which ri 4? (7.2.3) For a uniform member, A and E are the area of the cross-section and the elastic modulus, respectively. In addition, I is the length of the member. The second and fourth columns and rows of the stiffness matrix associated with the transverse displacement v are null since the truss member has axial deformation only. The plane truss structure consists of axial members in different orientations. For example, Fig. 7.2.2 shows that members a, b and c lie in three different directions. In order to assemble the stiffness matrices related to these truss members, we need200 ‘Truss Structures Chapter 7 y | AE " Ue p—___L________ Figure 7.2.1 Two-Dimensional Truss Element Figure 7.2.2. Triangular Truss to have the element degrees of freedom given in terms of the common reference axes. In other words, the element nodal displacements are expressed in terms of the fixed global coordinate system. Figure 7.2.3 shows a plane truss element oriented in an arbitrary angle 2 with respect to the horizontal axis 2. The figure shows two sets of nodal displacements. One set has nodal displacements along and perpendicular to the element axis (i.e. u and v) while the other set has the displacements in terms of the global reference axes (ie. & and 0). Because the element stiffness matrix Eq. (7.2.2) is expressed in terms of u and v, it should be transformed such that the stiffness matrix is expressed in terms of & and 0. To the end, we find the relationship between zy- and 29-coordinate systems. This relationship is called the coordinate transformation and the same relationship holds for the two sets of nodal displacements (i.e. u, v and i, 3). The relationship is uy ¢ 8 0 07 (mH ile jes e 0 Ol) a ee (1.2.4) v2 0 0 -s ¢ ta where ¢ = cosf and s = sin@. Let us rewrite Eq. (7.2.4) as {d°} = (THa"} (7.2.5)Section 7.2 Plane Truss 201 Figure 7.2.3 Generalized Two-Dimensional Truss Element To transform the element stiffness matrix from the zy-coordinate system to Zy-coordinate system, consider the concept of strain energy. The strain energy is expressed as v= hati} (728) in terms of the zy-coordinate system. If we substitute Eq. (7.2.5) into Eq. (7.26), we obtain : u = hey ore) (722) ‘The strain energy is now expressed in terms of the #j-coordinate system. Lyceytecer cz U= Haye} (7.2.8) in which [K¢] is the transformed element stiffness matrix in terms of the 24-coordinate system. The strain energy in Eq. (7.2.8) should be the same as that in Eq. (7.2.7) because strain energy is independent of the coordinate system. Equating Eq. (7.2.7) to Eq. (7.2.8) shows that (R= I (7.29) Substitution of Eqs (7.2.2) and (7.2.4) into Eq. (7.2.9) results in the transformed stiffness matrix food cs —c? -cs ze) AE} cs 5? ~cs —s? lege) 3 —c8 3? cs 8? (2.7.10) for the nodal degrees of freedom {a 0, t 2} (7.2.11) This element stiffness matrix can be assembled into the global matrix as usual for the shared nodal points.202 write ‘Truss Structures Example 7.2.1 Chapter 7 Let us compute element stiffness matrices for the truss structure shown in Fig. 7.2.2. The structure has three elements (a, b and c) and three nodes (1, 2 and 3). Let each truss member have the same material and geometric properties. In the following derivation, the superimposed (") is omitted for simplicity Member a: Let i=1 and j=: The element stiffness matrix is AE {k= 4 oocco Member b: Let i=2 and j=3. Then @ = s =sin B=“. The element stiffness matrix is 0.250 -0.433 AE | -0.433 0.750 a io {R= —F | 9.250 0.433 0.433 -0.750 Member c: Let i=1 and j=3, Then 8 8 0.250 AE | 0.433 ey 2 {Ky = tT | -0.250 —0.433 sin 8=%8. The element stiffness matrix is 0.433 0.750 —0.433 . Then 8=0 so that c = cos #=0 and s = sin 8=0. (7.2.12) oro coco 35 so that ¢ = cosf=-0.5 and —0.250 0,433 0.433 —0.750 0.250 ~0.433 (72.13) —0.433 0.750 = £ so that ¢ = cos§=0.5 and 0.250 -0.433 0.433 0.750 0.250 0.433 (7.2.14) 0.750 0.433 0.750 The element mass matrix for the plane truss member can be calculated using the same coordinate transformation. Using the kinetic energy expression, as similar to the strain energy expression for derivation of the element stiffness matrix, we can (M°] = ("MIT (7.2.15) Carrying out this matrix multiplication using the consistent mass matrix gives 2c? 2s 6 PAL | 2cs ¢ cs 2s? cs 2 2 es 2 css a (7.2.16) 2cs 2s?Section 7.3 Space Truss 203 The lumped mass matrix can be obtained similarly and shown below: 2 ce 0 0 pAl|cs 8? 0 0 OF 0) 0 ce ce 00 ca a2 (M‘] = (7.2.17) 7.3 Space Truss Development of the element stiffness matrix for the space truss member is similar to that for the plane truss member. The element stiffness matrix in terms of the global Cartesian coordinate system is obtained in the same way as given in Eq. (7.2.9). However, the sizes of both the transformation matrix and the element stiffness matrix in terms of the body coordinate system are 6x6 for the space truss member. Here, the body coordinate system denotes the coordinate system one of whose axes lies along the member direction. The stiffness matrix in terms of the body coordinate system is -1 1 0 AE| 0 7-— od (7.3.1) eoocoo eoooce coHnoo ooooce ccococe] 0 0 for the nodal degrees of freedom of {a }={u v1 wi tu v2 wa} (7.3.2) where u is the displacement along the z-axis as shown in Fig. 7.3.1. The transformation matrix between the two coordinate systems is given below: fo G0 0 0 re old (Nelo 0 0 a a 6 (73.3) 0 0 0 m m ns 60 0 a a G where {1 m 1} is the direction cosines of Z-axis with respect to zyz-coordinate system. Similarly, {€2 nz ¢2} and {€3 m3 Ca} are the direction cosines of g- and Z-axis with respect to zyz-coordinate system, respectively. Conducting the matrix manipulation yields G && fb - hbo —bks a(S 8 o8 “Be 8 ap rey — 13 &aks “hfs —S2bs —&3 Pee ee fel ee cen Ge (734) -a@ -¢ -66 && @ 66 -i& -66 -@ 4& &&204 ‘Truss Structures Chapter 7 Figure 7.3.1 Generalized Three-Dimensional Truss Element The corresponding clement degrees of freedom are {#}={th t th i t we} (7.3.5) ‘The consistent mass matrix for the space truss element is 27 ib Wk EF fb ffs Miko 26 Mobs bbe GF babs PAL | 215 rls 265 ibs baka 3 6 | & Ges Eile 2? sky riko ee aids Geka ee 28F Uae fla Gals GE res 268 while the lumped mass matrix is @& && 4 0 0 0 lie Bop oo 8 a eAl|éé Gf @ 0 0 0 MI="S]"o 0 0 ike ids (73.7) 0 0 0 && & &2f3 0 0 0 ce bee 7.4 MATLAB Application to Static Analysis The static analysis of a truss structure is to solve the following matrix equation: [K]{a} = {F} (7.4.1) where the system stiffness matrix [K] and the system force vector {F} are obtained by assembling each element: matrix and vector. This section shows some examples of MATLAB programs and m-files for static analyses of tao-dimensional ttiiss structures.Section 7.4 MATLAB Application to Static Analysis 205 @ azo.sint B=30x10" pet @ 4=0.4int E=30x10° psi 1000 w& Figure 7.4.1 Truss With Two Axial Members 4 Example 7.4.1 Figure 7.4.1 shows a simple truss structure made of two members. Each member has elastic modulus of E=30x 10° psi and cross- sectional areas are A=0.4 in? and A2=0.5 in? where subscript indicates the element number as shown in the figure. A 1000 Ib force is applied at the tip in the downward direction. Find the displacements and stresses of members. The MATLAB program and m-files are provided below. %— % Example 7.4.1 % to solve static 2-D truss structure % % Problem description % Find the deflection and stress of the truss made of two members % as shown in Fig. 7.4.1. % % Variable descriptions % k = element stiffness matrix % kk = system stiffness matrix % ff = system force vector % index = a vector containing system dofs associated with each element % gcoord = global coordinate matrix % disp = nodal displacement vector % elforce = element force vector % eldis) element nodal displacement % stress = stress vector for every element % elprop = element property matrix % nodes = nodal connectivity matrix for each element a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in bedof %-206 ‘Truss Structures Chapter 7 % eects! % control input data % number of elements % number of nodes per element % number of dofs per node % total number of nodes in system sdof=nnode*ndof; % total system dofs % 6 % nodal coordinates %— % x, y-coordinate of node 1 % x, y-coordinate of node 2 gcoord(3,1)=0.0; gcoord(3,2)=10.0; % x, y-coordinate of node 3 % ——— % material and geometric properties SS elprop(1,1)=30000000; % elastic modulus of Ist element elprop(1,2)=0.4; % cross-section of Ist element elprop(2,1)=30000000; % elastic modulus of 2nd element elprop(2,2)=0.5; % cross-section of 2nd element % % % nodal connectivity % nodes associated with element 1 % nodes associated with element 2 % Ast dof (horizontal displ) is constrained % whose described value is 0 % 2nd dof (vertical displ) is constrained % whose described value is 0 % Sth dof (horizontal displ) is constrained % whose described value is 0 % 6th dof (vertical displ) is constrained % whose described value is 0 % initialization to zero %&————___—_———__— ff=zeros(sdof,1); % system force vector kk=zeros(sdof,sdof); % system stiffness matrixSection 7.4 MATLAB Application to Static Analysis index=zeros(nnel*ndof,1); % index vector elforce=zeros(nnel*ndof,1); % element force vector eldisp=zeros(nnel*ndof,1); % element nodal displacement vector k=zeros(nnel*ndof,nnel*ndof); % element stiffness matrix stress=zeros(nel,1); % stress vector for every element % %—_______—_- % applied nodal force %—__—_____—. ff(4)=-1000; % 2nd node has 1000 Ib in downward direction % a % loop for elements %—_________— for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ast connected node for the (iel)-th element nd(2)=nodes(iel,2); % 2nd connected node for the (iel)-th element % xl=gcoord(nd(1),1); yl=gcoord(nd(1),2); % coordinate of ist node x2=gcoord(nd(2),1); y2=gcoord(nd(2),2); % coordinate of 2nd node % leng=sqrt((x2-x1)*2+(y2-y1)*2); % element length % if (x2-x1)==0; beta=2*atan(1); % angle between local and global axes else beta=atan((y2-y1)/(x2-x1)); % extract elastic modulus aren=elprop( el; 2); % extract cross-sectional area index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=fetruss2(el,Jeng,area,0,beta,1); % compute element matrix % kk=feasmbl1 (kk,k,index); % assemble into system matrix % end % %——_—___—__—- % apply constraints and solve the matrix Cena canna [kk,ff]=feaplyc2(kk,ff,bedof, beval); % apply boundary conditions % disp=kk\ff; % solve matrix equation for nodal displacements % 207208 Truss Structures %—_—_____—— % post computation for stress calculation for iel=1:nel % nd(1)=nodes(iel,1); nd(2)=nodes(iel,2); % x1=gcoord(nd(1),1); yl=gcoord(nd(1),2); coord (nd(2),1); y2=gcoord(nd(2),2); % leng=sqrt((x2-x1)"2+(y2-y1)"2); % if (x2-x1)==0; beta=2*atan(1); else beta=atan((y2-y1)/(x2-x1)); end % el=elprop(iel,1); area=elprop(iel,2); % index=feeldof(nd,nnel,ndof); % k=fetruss2(el,leng,area,0,beta,1); % for nel*ndof) eldisp(i)=disp(index(i)); end % elforce=k*eldisp; stress(iel)=sqrt(elforce(1)3-+elforce(2)2)/area; % if ((x2-x1)*elforce(3)) < 0; stress(iel)=-stress(iel); end % end % — % print fem solutions (ee of; displ=[num’ disp] % numm=! nel; stresses=[numm’ stress] % Chapter 7 % loop for the total number of elements % Ast connected node for the (iel)-th element % 2nd connected node for the (iel)-th element % coordinate of 1st node % coordinate of 2nd node % element length % angle between local and global axes % extract elastic modulus % extract cross-sectional area % extract system dofs for the element % compute element matrix % extract displacements associated with % (iel)-th element % element force vector % stress % check if tension or compression % print displacements % print stressesSection 7.4 MATLAB Application to Static Analysis 209 %— function {k,m]=fetruss2(el,leng,area,tho, beta,ipt) % % Purpose: % Stiffness and mass matrices for the 2-D truss element % nodal dof { wl v-1u.2 v.2} % % Synopsis: % (k,m]=fetruss2(cl,leng,area,tho,beta,ipt) % % Variable Description: % k - clement stiffness matrix (size of 4x4) % m - element mass matrix (size of 4x4) % el - elastic modulus % Jeng - element length % area - area of truss cross-section % tho - mass density (mass per unit volume) % beta - angle between the local and global axes % Positive if local axis is in cew direction from % the global axis % ipt = 1 - consistent mass matrix % ipt = 2 - lumped mass matrix % % % stiffness matrix % c=cos(beta); s=sin(beta); k= (area*el/leng)*[ ctc cts -s s*s -cte -cts acts -sts % % consistent mass matrix % if ipt==1 % m=(rho*area*leng/6)*[2 01 07 2) 0 10 2 0 1 0 % % lumped mass matrix % else %210 ‘Truss Structures Chapter 7 % end % % ‘The results from the finite element analysis are given below. The minus sign in the stress indicates compressive stress. displ = dofs displacement 1.0000 0.0000 % horizontal displ. of node 1 2.0000 0.0000 % vertical displ. of node 1 3.0000 -0.0008 % horizontal displ. of node 2 4.0000 -0.0027 % vertical displ. of node 2 5.0000 0.0000 % horizontal displ. of node 3 6.0000 0.0000 % vertical displ. of node 3 stresses = element stress. 1.00000 -2500. % compressive stress for element 1 2.00000 2828. % tensile stress for element 2 4 Example 7.4.2 Find the stresses of the truss structure shown in Fig. 7.4.2. All members have elastic modulus of 200 GPa and cross-sectional area of 2.5 x 10-3 m?.Section 7.4 MATLAB Application to Static Analysis 211 %- % Example 7.4.2 % to solve static 2-D truss structure % % Problem description % Find the deflection and stress of the truss made of two members % as shown in Fig. 7.4.2. % % Variable descriptions % k = element stiffness matrix % kk = system stiffness matrix % ff = system force vector % index = a vector containing system dofs associated with each element % gcoord = global coordinate matrix % disp = nodal displacement vector % elforce = element force vector % eldisp = clement nodal displacement % stress = stress vector for every element % prop = material and geometric property matrix % nodes = nodal connectivity matrix for each element % bedof = a vector containing dofs associated with boundary conditions vector containing boundary condition values associated with % the dofs in bedof % number of elements % number of nodes per element % number of dofs per node nnode=6; % total number of nodes in system sdof=nnode*ndof; % total system dofs % %— % nodal coordinates % gcoord(1,, gcoord(2,. gcoord(3,1 gcoord(4,, gcoord (5, gcoord(6, % %— ee % material and geometric properties ——— prop(1)=200e9; % elastic modulus212 ‘Truss Structures Chapter 7 prop(2)=0.0025; % cross-sectional area % a % nodal connectivity nodes(4,1)=2; nodes(5,1)=3; nodes(5,2) nodes(6,1)=3; nodes(6,2)=! nodes(7,1)=4; nodes(7,2): nodes(8,1)=4; nodes(8,2): nodes(9,1)=5; nodes(9,2): % %— _—_____—. % applied constraints —— bedof(1): % Ast dof (horizontal displ) is constrained beval(1) % whose described value is 0 bedof(2): % 2nd dof (vertical displ) is constrained beval(2)=0; % whose described value is 0 bedof(3)=12; % 12th dof (vertical displ) is constrained beval(3)=0; % whose described value is 0 % oe % initialization to zero oe Ss ff=zeros(sdof,1); % system force vector kk=zeros(sdof,sdof); % system stiffness matrix index=zeros(nnel*ndof,1); % index vector elforce=zeros(nnel*ndof,1); % element force vector eldisp=zeros(nnel*ndof,1); % element nodal displacement vector k=zeros(nnel*ndof,nnel*ndof); % element stiffness matrix stress=zeros(nel,1); % stress vector for every element %— % applied nodal force % 4th node has 600 N in downward direction % 5th node has 200 N in r-h.s. direction % loop for elements %— for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ist connected node for the (iel)-th elementSection 7.4 MATLAB Application to Static Analysis nd(2)=nodes(iel,2); % 2nd connected node for the (iel)-th element % x1=gcoord(nd(1),1); yl=gcoord(nd(1),2); % coordinate of Ist node x2=gcoord(nd(2),1); y2=gcoord(nd(2),2); % coordinate of 2nd node % leng=sqrt((x2-x1)"2+(y2-y1)°2); % element length % if (x2x ; eet, % angle between local and global axes else beta=atan((y2-y1)/(x2-x1)); end % el=prop(1); % extract elastic modulus area=prop(2); % extract cross-sectional area % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=fetruss2(el,leng,area,0,beta,1); % compute element matrix % kk=feasmbl1(kk,k,index); % assemble into system matrix % end % ———— % apply constraints and solve the matrix [kk,ft]=feaplyc2(kk,ff, bedof, bcval); % apply boundary conditions % disp=kk\ ff; % solve matrix equation to find nodal displacements % % % post computation for stress calculation 6——— $$ $$$ __________ % for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ast connected node for the (iel)-th element nd(2)=nodes(iel,2); % 2nd connected node for the (iel)-th element % x1=geoord(nd(1),1); yl=gcoord(nd(1),2); % coordinate of Ist node x2=gcoord(nd(2),1); y2=gcoord(nd(2),2); % coordinate of 2nd node % leng=sqrt((x2-x1)"2+(y2-y1)°2); % element length % if (x2-x1 ect, % angle between local and global axes else 213214 Truss Structures beta=atan((y2-y1)/(x2-x1)); end % Chapter 7 el=prop(1); % extract elastic modulus area=prop(2); % extract cross-sectional area index=feeldof(nd,nnel,ndof); % extract system dofs for the element % k=fetruss2(el,Jeng,area,0,beta,1); % compute element matrix % for eldisp(i end % (nnel*ndof) ‘% extract displacements associated with isp(index(i)); % (iel)-th element elforce=k*eldisp; % element force vector stress(iel)=sqrt(clforce(1)2+elforce(2)3)/area; % % stress if ((x2-x1)*elforce(3)) < 0; % check if tension or compression stress(iel)=-stress(iel); end % end % %— % print fem solutions The nodal displacements and stresses of members are shown below. displ = dofs displacement 1.0000 0.0000e-0 2.0000 0.0000e-5 3.0000 0.3200e-5 4.0000 -1.5700e-5 5.0000 0.8650e-5 6.0000 —-1.5700e-5 7.0000 0,6400e-5 8.0000 —--2.2867e-5 9.0000 0,5450e-5 10.000 -2.0167e-5 % print displacements % print stressesSection 7.5 0 MATLAB Application to Eigenvalue Analysis ee @ set'ne — B=2006Pa i. p=7860kg/m Figure 7.5.1 Finite Element Discretization 11.000 1.1200e-5 12.000 ——-0.0000e-5 stresses = element stress 1.0000 160000 2.0000 100000 3.0000 000000 4.0000 160000 5.0001 100000 6.0001 -160000 7.0001 180000 8.0001 240000 9.0001 — -300000 215 7.5 MATLAB Application to Eigenvalue Analysis Once the system mass and stiffnes matrices are computed for the truss structure, the matrix equation becomes [M]{i} + [A]{u} = 0 (75.1) In order to compute natural frequencies of the structure, we assume a harmonic motion for the displacement. The resultant equation is the eigenvalue problem given as ([K] ~ w?[M]){a} = 0 (7.5.2) where w is the circular natural frequency and {i} is the vector for mode shape. 4 Example 7.5.1 Determine the natural frequency of a free bar using the finite element method. The bar is shown in Fig. 7.5.1 and it has elastic modulus of 200 GPa, cross-sectional area of 0.001 m?, and density of 7860 Kg/m°.216 ‘Truss Structures Chapter 7 % % Example 7.5.1 % to solve natural frequency of 1-D bar structure % % Problem description % Find the natural frequency of a bar structure % as shown in Fig. 7.5.1. % % Variable descriptions % k = element stiffness matrix % m = element mass matrix % kk = system stiffness matrix % mm = system mass vector % index = a vector containing system dofs associated with each element % gcoord = global coordinate matrix % prop = element property matrix % nodes = nodal connectivity matrix for each element % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in bedof % number of elements % number of nodes per element % number of dofs per node % total number of nodes in system sdof=nnode*ndof; % total system dofs % ————————_—_———— % nodal coordinates %— gcoord(1,1 geoord(2,1)=1. gcoord(3,1)=2.0; gcoord(4,1)=3. gcoord(5,1)=4.0; % % % material and geometric properties —— prop(1)=200e9; % elastic modulus % cross-sectional area % densitySection 7.5 % nodal connectivity %e nodes(1,1)=1; nodes(1,2): nodes(2,1)=2; nodes(2,2)=3; nodes(3,1)=3; nodes(3,2) nodes(4,1)=4; nodes(4,2) % % % initialization to zero %— kk=zeros(sdof,sdof); mm=zeros(sdof,sdof); index=zeros(nnel*ndof,1); % %— % loop for elements %— MATLAB Application to Eigenvalue Analysis % system stiffness matrix % system mass matrix % index vector for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); nd(2)=nodes(iel,2); % Ast connected node for the (iel)-th element % 2nd connected node for the (iel)-th element Teng=(x2-x1); % coordinate of 1st node % coordinate of 2nd node % element length % el=prop(1); % extract elastic modulus area=prop(2); % extract cross-sectional area tho=prop(3); % extract mass density % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % ipt=1; % flag for consistent mass matrix [k,m]=fetruss1 (el,leng,area,tho,ipt); % Kk=feasmbl1(kk,k,index); mm=feasmbl1 (mm,m,index); % end % es % solve for eigenvalues ee faol=cig(Ikk,mm faol=sqrt(fsol); % % % element matrix % assemble system stiffness matrix % assemble system mass matrix 217218 ‘Truss Structures Chapter 7 % print fem solutions ae num=l:1:sdof; freqcy=[num’ fsol] % print natural frequency % % function [k,m]=fetruss1 (el,leng,area,rho,ipt) % Purpose: % Stiffness and mass matrices for the 1-D truss element % nodal dof { u-1 u-2 } % % Synopsis: % (k,m]=fetruss1 (el,leng,area,tho,ipt) % % Variable Description: % k - element stiffness matrix (size of 4x4) % m - element mass matrix (size of 4x4) % el - elastic modulus % leng - element length % area - area of truss cross-section % cho - mass density (mass per unit volume) % ipt = 1 - consistent mass matrix % ipt = 2 - lumped mass matrix %e % % stiffness matrix % k= (area*el/leng)*[ 1 ok % % consistent mass matrix % if ipt: % m=(rho*area*leng/6)*[ 2 1 2}; % % lumped mass matrix % else % m=(rho*area*leng/2)*[ 1 0 % endSection 7.5 MATLAB Application to Eigenvalue Analysis 219 % % The natural frequencies are computed from the finite clement analysis and compared to the exact solution. freqcy = mode nat. freq. 1.00 0.0000 % exact 0.0000 2.00 4060.0 % exact 3962.0 3.00 8737.0 % exact 7924.0 4.00 14198. % exact 11895. 5.00 17474. % exact 15847, # Example 7.5.2 We want to find the natural frequency of the truss structure shown in Fig. 7.4.2. Each member has density of 7860 Kg/m°. —————— % Example 7.5.2 % to solve natural frequency of 2-D truss structure % % Problem description % Find the natural frequency of a truss structure % as shown in Fig. 7.4.2. % % Variable descriptions % k = element stiffness matrix % m = element mass matrix % kk = system stiffness matrix % mm = system mass vector % index = a vector containing system dofs associated with each element % gcoord = global coordinate matrix % prop = element property matrix % nodes = nodal connectivity matrix for each element % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in *bcdof” % % a % control input data % number of elements % number of nodes per element % number of dofs per node220 ‘Truss Structures Chapter 7 mnode=6; % total number of nodes in system sdof=nnode*ndof, % total system dofs % a % nodal coordinates gcoord(5,1)=8.0; gcoord(5,2)=3.0; gcoord(6,1)=12.; gcoord (6,2): % %— ————— % material and geometric properties K-——_ ——_— prop(1)=200e9; % elastic modulus prop(2)=0.0025; % cross-sectional area prop(3)=7860; % density % — % nodal connectivity nodes(2,1)=1; nodes(2,2)=: nodes(9,1)=5; nodes(9,2)= % %— % applied constraints % Ast dof (horizontal displ) is constrained % whose described value is 0 % 2nd dof (vertical displ) is constrained % whose described value is 0 % 12th dof (vertical displ) is constrained % whose described value is 0 % initialization to zero %— a kk=zeros(sdof,sdof); % system stiffness matrix mm=zeros(sdof,sdof); % system mass matrixSection 7.5 MATLAB Application to Eigenvalue Analysis index=zeros(nnel*ndof,1); % index vector % ——— % loop for elements %—__—____— for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ast connected node for the (iel)-th element nd(2)=nodes(iel,2); % 2nd connected node for the (iel)-th element % xl=gcoord(nd(1),1); yl=geoord(nd(1),2); % coordinate of Ist node x2=gcoord(nd(2),1); y2=gcoord(nd(2),2); % coordinate of 2nd node % leng=saqrt((x2-x1)*2+(y2-y1)"2); % element length % if (x2-x1)==0; beta=2*atan(1); % angle between local and global axes else beta=atan((y2-y1)/(x2-x1)); end % el=prop(1); ‘% extract elastic modulus area=prop(2); % extract cross-sectional area tho=prop(3); % extract mass density index=feeldof(nd,nnel,ndof); % extract system dofs for the element % ip’ % flag for consistent mass matrix [k,m]=fetruss2(el,leng,area,rho,beta,ipt); % element matrix % kk=feasmbl1(kk,k,index); % assemble system stiffness matrix mm=feasmbl1(mm,m,index); % assemble system mass matrix % end % Ee % apply constraints and solve %& [kk,mm]=feaplycs(kk,mm,bedof); % apply the boundary conditions % fsol=eig(kk,mm); fsol=sqrt (fsol); % %—_——___—_——_—— % print fem solutions sdof; freqcy=[num’ fsol] % print natural frequency 221222 Truss Structures Chapter 7 function [kk,mm]=feaplycs(kk,mm,bcdof) %— ‘% Purpose: % Apply constraints to eigenvalue matrix equation % (kk}x=lambda[mm]x % % Synopsis: % [kk,mm]=feaplycs(kk,mm,bedof) % % Variable Description: % kk - system stiffness matrix before applying constraints % mm - system mass matrix before applying constraints % bedof - a vector containing constrained d.o.f % ‘% Notes: % This program does not reduce the matrix size depending on % the number of constraints. Instead the system matrix size % is preserved regardless of constraints. As a result, the % matrix obtained after applying the constraints contain fictitious % zero eigenvalues as many as the number of constraints in % addition to actual eigenvalues. Users neglect the zero % fictitious eigenvalues from the results. ————————“——_——_ % n=length(bedof); sdof=size(kk); % for i=1:n c=bedof(i); mm(j,c)=0; end % mm(c,c)=1; end % % The first five natural frequencies of the truss structure are provided below.Section 7.6 MATLAB Application to Transient Analysis 223 Ist frequency = 240.9 rad/s Ind frequency = 467.9 rad/s 3rd frequency = 739.8 rad/s 4th frequency = 1243. rad/s Sth frequency = 1633. rad/s 7.6 MATLAB Application to Transient Analysis The dynamic equation of motion for the truss structure is [Ma + La}! = (76.1) with prescribed initial conditions which are usually initial displacements and initial velocities. We apply the central difference technique for time integration of Eq. (7.6.1). The details of the techniques are described in Sec. 8.11 and are omitted here. In particular, the summed form of central difference technique is used for the following examples. & Example 7.6.1 A bar is fixed at the left end and it is subjected to a step function of magnitude of 200 N (see Fig. 7.5.1). The bar has elastic modulus of 200 GPa, cross-sectional area of 0.001 m?, and density of 7860 Kg/m*. It is initially at rest. The MATLAB program is shown below. % % Example 7.6.1 % to solve transient response of 1-D bar structure % % Problem description % Find the dynamic behavior of a bar structure, % as shown in Fig. 7.5.1, subjected to a step % force function at the right end. % % Variable descriptions % k = clement stiffness matrix % element mass matrix % kk = system stiffness matrix % mm = system mass vector % ff = system force vector % index = a vector containing system dofs associated with each element % gcoord = global coordinate matrix % prop = element property matrix nodal connectivity matrix for each element a vector containing dofs associated with boundary conditions224 ‘Truss Structures Chapter 7 % beval = a vector containing boundary condition values associated with % the dofs in bedof — % %— % control input data % number of elements % number of nodes per element % number of dofs per node % total number of nodes in system % total system dofs % time step size % initial time % final time nt=fix((t£ti)/dt); % number of time steps % Ip % nodal coordinates gcoord(1, gcoord(2, gcoord(3, gcoord(4, gcoord(5, gcoord(6, gcoord(7, gcoord(8, gcoord(9, gcoord(10,1)=9, gcoord(11,1)=10.0; % material and geometric properties La prop(1)=200e9; % elastic modulus prop(2)=0.001; % cross-sectional area prop(3)=7860; % density % &— % nodal connectivitySection 7.6 MATLAB Application to Transient Analysis nodes(10,1)=10; nodes(10,2)=11; % %~—— % applied constraints % nbe=1; % number of constraints bedof(1)=1; % Ast dof is constrained % ~%— % initialization to zero oo ae kk=zeros(sdof,sdo! % system stiffness matrix mm=zeros(sdof,sdof); % system mass matrix fl=zeros(sdof,1); % system force vector index=zeros(nnel*ndof,1); % index vector acc=zeros(sdof,nt); % acceleration matrix vel=zeros(sdof,nt); % velocity matrix disp=zeros(sdof,nt); % displacement matrix % %— % loop for elements (eee for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ast connected node for the (iel)-th element nd(2)=nodes(iel,2); % 2nd connected node for the (iel)-th element % coord (nd(1),1); % coordinate of Ist node x2=gcoord(nd(2),1); % coordinate of Ind node % leng=(x2-x1); % clement length % el=prop(1); % extract elastic modulus area=prop(2); % extract cross-sectional area tho=prop(3); % extract mass density index=feeldof(nd,nnel,ndof); ‘% extract system dofs for the element. % ipt=1; % flag for consistent mass matrix [k,m]=fetruss1 (el,leng,area,tho,ipt); % element matrix % kk=feasmbl1(kk,k,index); % assemble system stiffness matrix mim=feasmbli(mm,m,index); % assemble system mass matrix % end 225226 ‘Truss Structures Chapter 7 % % initial condi ee vel(:,1)=2eros(sdof,1); % initial zero velocity disp(:,1)=zeros(sdof,1); % initial zero displacement % fi(11)=200; % step force at node 11 % ¢—$ $$$ % central difference scheme for time integration a mm=iav(mm); % invert the mass matrix % for it=1:nt % ace(;,it)=mm*(ff-kk*disp(;,it)); % compute acceleration % for i=1:nbe ibc=bedof(i); % apply constraints ace(ibe,it)=0; end % compute velocity % compute displacement acc(:,nt-4+1)=mm*(ff-kk*disp(:,nt+1)); % acceleration at last step % time=O0:dt:nt*dt; plot(time,disp(11,:)) xlabel(*Time(seconds)’) ylabel("Tip displ. (m)’) % %%—— ‘The tip displacement at the right end (i.e. node 11) is plotted in Fig. 7.6.1. as a function of time. As expected, the displacement has oscillation about the static displacement. & Example 7.6.2 Find the transient response of the truss structure as shown in Fig. 7.4.2. The structure has the same geometric and material data as those given Example 7.5.2. However, the load is applied at node 5 in the upward direction as a step function. The load magnitude is 200 N. The response of the same node where the load is applied is plotted in Fig. 7.6.2.Section 7.6 MATLAB Application to Transient Analysis r Jo Pod 4 ify a a a “Tinatecot) Figure 7.6.2 Time History of Node 5 a % Example 7.6.2 ‘% to solve transient response of 2-D truss structure % % Problem description % Find the dynamic behavior of a truss structure, % as shown in Fig. 7.4.2, subjected to a step % force function at node 5 in the upward direction. % % Variable descriptions % k = element stiffness matrix % m = element mass matrix % kk = system stiffness matrix % mm = system mass vector % ff = system force vector 227228 Truss Structures Chapter 7 % index = a vector containing system dofs associated with each element % gcoord = global coordinate matrix % prop = element property matrix % nodes = nodal connectivity matrix for each element % bedof = a vector containing dofs associated with boundary conditions % beval vector containing boundary condition values associated with % the dofs in bedof 6 $$$ ______________-% es % control input data % number of elements % number of nodes per element % number of dofs per node % total number of nodes in system sdof=nnode*ndof; % total system dofs dt=0.0005; % time step size i=0; % initial and final times nt=fix((t£ti)/dt); % number of time steps % % % nodal coordinates % elastic modulus % cross-sectional area % density % nodal connectivity %— nodes(1,1)=1; nodes(1,2Section 7.6 MATLAB Application to Transient Analysis nodes(9,1)=5; nodes(9,2)=6; % % % applied constraints bedof(3)=12; beval(3)=0; % ——————————— % initialization to zero % kk=zeros(sdof,sdof); mm=zeros(sdof,sdof); ff=zeros(sdof,1); index=zeros(nnel*ndof,1); % number of constraints % Ast dof (horizontal displ) is constrained % whose described value is 0 % 2nd dof (vertical displ) is constrained % whose described value is 0 % 12th dof (horizontal displ) is constrained % whose described value is 0 % system stiffness matrix % system mass matrix % system force vector % dots index vector acc=zeros(sdof,nt); % acceleration matrix vel=zeros(sdof,nt); % velocity matrix disp=zeros(sdof,nt); % displacement matrix % % % loop for elements ‘% loop for the total number of elements % \st connected node for the (iel)-th element % 2nd connected node for the (iel)-th element % coordinate of ist node % coordinate of 2nd node leng=sqrt((x2-x1)"2+(y2-yl)*2); % element length % if (x2-x1)==0; beta=2*atan(1); % angle between local and global axes else beta=atan((y2-y1)/(x2-x1)); end % el=prop(1); % extract elastic modulus area=prop(2); % extract cross-sectional area rho=prop(3); % extract mass density % index=feeldof(nd,nnel,ndof); % extract system dofs for the element 229230 ‘Truss Structures Chapter 7 % ipt=1; % flag for consistent mass matrix [k,m]=fetruss2(el,leng,area,rho, beta,ipt); % element matrix kk=feasmbl1 (kk,k,index); % assemble system stiffness matrix mm=feasmbli (mm,m,index); % assemble system mass matrix % end % %— % initial condition % vel( eros(sdof,1 % initial zero velocity disp(:,1)=zeros(sdof,1); % initial zero displacement % 1(10)=200; % step force at 10th dof % %— % central difference scheme for time integration ——————— mm=inv(mm); % invert the mass matrix % for it=I:nt % acc(:,it)=mm*(ff-kk*disp(,it)); % compute acceleration % for i=1:nbe ibe=bedof(i); % apply constraints acc(ibe,it)=0; end % vel(:,it-4+1)=vel(:,it)-+acc(:,it)*dt; % compute velocity disp(;,it-+1)=disp(:,it)+-vel(:,it-+1)*dt; % compute displacement % end % ace(:nt-+1)=mm*(ff-kk*disp(:;nt-+1)); % acceleration at last step % time=0:dt:nt*dt; plot (time,disp(10,;)) % displacement. plot xlabel(’Time(seconds)’) ylabel(’Tip displ. (m)’) % %TA 72 73 74 75 76 LT 78 79 7.10 TAL Problems 231 Problems (a) Develop the element stiffness matrix for the one-dimensional axial rod using quadratic shape functions. (b) Apply the stiffness matrix to solve an axial member whose one end is fixed and the other end is subjected to a force P. The member has elastic modulus E and cross-sectional area A, respectively. Use one quadratic element to model the axial member. (c) Compare the nodal displacement at the center of the member to the end displacement. A telescope shape of axial member (see Fig. P7.2) is modeled using a single linear element. Derive the element stiffness matrix. A taper shape of axial member (Fig. P7.3) is modeled as a single linear element. Derive the element stiffness matrix. Develop the element mass matrix for the one-dimensional axial member using quadratic shape functions. Develop the element mass matrix for Prob. 7.2. Develop the element mass matrix for Prob. 7.3. For the truss structure shown in Fig. P7.7, derive the finite element matrix equation using two elements. Find the displacements and stresses in the member. One-dimensional wave equation for an axial member is given as Guo _ ot On? The second order equation can be rewritten as a Pu era using two first order equations in time. These two equations are solved using one linear finite element and the backward difference method for time derivative. The member is initially displaced such that u(z,0)=0.001z. The left end of the member is held fixed all the time while the right end is released at time 0 from the initial displacement. Find the displacement u and velocity v at the right end at time t=1 sec. using a time step size At=1. Redo Prob. 7.8 using the central finite difference method for time derivative. Find the critical time step size for stability. Solve the truss structure shown in Fig. P7.10 using the computer programs. Compare the finite element solution to the analytical solution obtained from statics. Obtain the natural frequencies of the structures in Fig, P7.10 using the computer programs232 ‘Truss Structures Chapter 7 Figure P7.2 Problem 7.2 E, Asaxtle ++} Figure P7.3 Problem 7.3 2m Figure P7.7 Problem 7.7 800 N 250 N 1.2m e mM. tm tm. tm. Im B=70GPA Figure P7.10 Problem 7.10 7.12 If the structure in Fig. P7.10 is initially at rest and the forces are appliedsuddenly at time 0, determine the dynamic response of the structure using the computer programs.CHAPTER EIGHT BEAM AND FRAME STRUCTURES 8.1 Euler-Bernoulli Beam The Euler-Bernoulli equation for beam bending is Ov ay 5a t (23) = 4(z,¢) (8.1.1) where v(z,£) is the transverse displacement of the beam, p is mass density per volume, EI is the beam rigidity, q(z, 7) is the externally applied pressure loading, and ¢ and z indicate time and the spatial axis along the beam axis. We apply one of the methods of weighted residual, Galerkin’s method, to the beam equation, Eq. (8.1.1) to develop the finite element formulation and the corresponding matrix equations. ‘The averaged weighted residual of Eq. (8.1.1) is Le 2, 2 2, I =f (ae + we") - 1) ode =0 (8.1.2) where L is the length of the beam and w is a test function. The weak formulation of Eq. (8.1.2) is obtained from integrations by parts twice for the second term of the equation. In addition, discretization of the beam into a number of finite elements gives dv dw aw xf, pgudes BS Side f aude] + [rm] =0 (8.1.3) where V = £1(8*v/dz°) is the shear force, M = E'1(8*v/8z?) is the bending moment, Q° is an element domain and n is the number of elements for the beam. We consider shape functions for spatial interpolation of the transverse deflection, v, in terms of nodal variables. Interpolation in terms of the time domain will be discussed later. To this end, we consider an element which has two nodes, one at each 235236 Beam and Frame Structures Chapter 8 ‘| “| ws as ——————— | 2,=0 t= Figure 8.1.1 Two-Noded Beam Element end, as shown in Fig. 8.1.1. The deformation of a beam must have continuous slope as well as continuous deflection at any two neighboring beam elements. To satisfy this continuity requirement each node has both deflection, v; and slope, 0;, as nodal variables. In this case, any two neighboring beam elements have common deflection and slope at the shared nodal point. This satisfies the continuity of both deflection and slop. The Euler-Bernoulli beam equation is based on the assumption that the plane normal to the neutral axis before deformation remains normal to the neutral axis after deformation (see Fig. 8.1.2). This assumption denotes 8 = du/dz (i.e. slope is the first derivative of deflection in terms of x), Because there are four nodal variables for the beam element, we assume a cubic polynomial function for v(x) v(x) = eo + ere + cz? + cgx* (8.1.4) From the assumption for the Euler-Bernoulli beam, slope is computed from Eq. (8.1.4) O(x) = cy + cox + 3cgzx? (8.1.5) Evaluation of deflection and slope at both nodes yields (0) =co= 11 (0) = e1 = U(1) = cg + e4! + cpl? + esl = v9 (8.1.6) (1) = c1 + 2cal + 3egl? = 02 Solving Eq. (8.1.6) for ¢: in terms of the nodal variables u; and 6; and substituting the results into Eq. (8.1.4) give v(z) = Hi(z)vr + Ho(2)01 + Ha(2)u2 + Ha(2)92 (8.1.7) where (8.1.8)Section 8.1 Euler-Bernoulli Beam 237 i Figure 8.1.2 Euler-Bernonlli Beam Figure 8.1.3 Hermitian Beam Element The functions H;(2) are called Hermitian shape functions and shown in Fig. 8.1.3. The Hermitian shape functions are of C-type which means they make both v and 4v/ Oz continuous between two neighboring elements. Further discussion on C-type continuity, where n is an integer equal to or larger than zero, is provided in Chapter 12, Application of Hermitian shape functions and Galerkin’s method to the second term of Eq. (8.1.3) results in the stiffness matrix of the beam element. That is, 1 tx] = [ weraiee (19) ; where (B= (ay my ag HY) (e.110) and the corresponding element nodal degrees of freedom is {d°} = {v; 61 v2 A2}”. In Eq. (8.1.10), double prime denotes the second derivative of the function and J in Eq. (8.1.9) is the length of a beam element. Assuming the beam rigidity EJ is constant238 Beam and Frame Structures Chapter 8 within the element, the element stiffness matrix is ne 6l -12 61 TEL | Gl 42 6) on Wleae | 20 8 (1.11) 6 2-61 ~4P In case the beam rigidity is not constant within a beam element, the integral in Eq. (8.1.9) must be evaluated including EJ as a function of z. If the beam element is relatively short, for example in a refined mesh, the average value of EI for the element may be used with Bq. (8.1.11) for a simple and reasonable approximation. The third term in Eq. (8.1.3) results in the element force vector. For a generally distributed pressure loading, we need to compute Hy 1 f= [ a) a dz (8.1.12) Ha in which {F*} is the element force vector. If we have a uniform pressure load go within the element, the element force vector becomes (ih 6l a A 9) P (P) =a [ Bi hde= 8 a (8.1.13) : e Another common load type is a concentrated force within a beam element as shown in Fig. 8.1.4. In this case, the element force vector is , mh Hy (20) {FI} = / Po6(z — 20) 2 dz = Py ey (8.1.14) Hy H,(z0) where Po is the concentrated force applied at x = 2 and 6(z — 29) is the Dirac delta function. Element force vectors for some other cases are summarized in Fig. 8.1.4. The last term in Eq. (8.1.3) is the boundary conditions of shear force and bending moment at the two boundary points, z = 0 and z = L, of the beam. If these boundary conditions are known, the known shear force and/or bending moment are included in the system force vector at the two boundary nodes. Otherwise, they remain as unknowns. However, deflection and/or slope are known as geometric boundary conditions for this case. For static bending analyses of beams, the first term in Eq. (8.1.3) which is the inertia force term is neglected. As a result, assembling the element stiffness matrices and vectors results in the system matrix equation given below: {K] {4} = {F} (8.1.15)Section 8.1 Euler-Bernoulli Beam 239 [F°] He 16 oy ie UP T=P, Ll) Hy) Ha) Hy] (FEM, [4x (mrt) UC-tap30,) baat) ab (90521) (F*|-as zat sil. Figure 8.1.4 Element Force Vectors for Various Pressure Loads Given boundary conditions are applied to Eq. (8.1.15) and the matrix equations are solved for the unknown nodal variables, deflections and slopes. An example is given in Example 8.1.1. & Example 8.1.1 Solve a cantilever beam subjected to a tip load as shown in Fig. 8.1.5. Let us use one Hermitian beam element to solve the tip deflection. In this case, the element stiffness matrix is the same as the system stiffness matrix. The resultant element matrix equations 12 6b ~i2 6h] (% Yi EI| 6L 42? ~6L 21? alo jm T |-12 -62 12 -6L|)uf(7~)-P (8.1.16) 6b 2? ~6L 417) la, 0 In this equation, Vj and My are unknown reactions at the clamped support. The minus sign indicates the tip force is applied to the opposite direction to the deflection. The boundary conditions prescribed in terms of nodal variables are v1 = 0 and 6; = 0. Applying these conditions to Eq. (8.1.16) as described in240 Beam and Frame Structures Chapter 8 P Figure 8.1.5 Cantilever Beam the previous chapter and solving the resultant equation yield PLS oan (8.1.17) v= which is the exact solution for the cantilever beam. The reason the finite element analysis with one element results in the exact solution is the following. The Hermitian shape functions are based on a general cubic polynomial as seen in Eq. (8.1.4). The exact solution for the cantilever beam with a tip force is also a cubic function. As a result, the Hermitian shape functions can result in the exact solution. f For dynamic analyses of beams, the inertia force needs to be included. In this case, the transverse deflection is a function of z and ¢. The deflection is interpolated within a beam element as given below: o(z, t) = Hi(x)ui(t) + Ho(x)01(t) + Ha(x)vo(t) + Ha(x)O2(¢) (8.1.18) Equation (8.1.18) states that the shape functions are used to interpolate the deflection in terms of the spatial domain and the nodal variables are functions of time. The first term in Eq. (8.1.3) becomes ff otnitunaetéy (8.1.19) b where [H] = [Fy Hy He Ha) (8.1.20) and superimposed dot denotes temporal derivative. From Eq. (8.1.19), the element mass matrix becomes 1 [Mw] = f pA" [Haz 156 22154-1381 _pAl| 221 4P 13l -37 420} 54 131 156 -22I (8.1.21) -131 —3 221 4Section 8.1 Euler-Bernoulli Beam 241 The mass matrix in Eq. (8.1.21) is called the consistent mass matrix. Archer [11,12] is credited for the first development of the consistent mass matrix. Adding the components in the mass matrix, which are associated with only the displacement nodal variables (i.e. v; and v2), yields pAl, total mass of the beam element. The beam element conserves the mass in terms of its translational degrees of freedom. In the dynamic analysis, the system mass matrix is usually required to be in- verted. From this aspect, a diagonalized mass matrix has a computational advantage. One such matrix is 1 ey. PAL | 0 IM}=]5 (8.1.22) This matrix is called the lumped mass matrix, which was developed earlier than the consistent mass matrix. This matrix has half of the element mass at each translational nodal degree of freedom. Both mass matrices conserve the mass associated with their translational degrees of freedom. Another way to develop a diagonalized mass matrix from the consistent mass matrix is summarized below[13}. 1. Add the diagonal components of the consistent mass matrix associated with the translational degrees of freedom, i.e. the first and third diagonal components for the present beam element. The sum is called «. 2, Divide the diagonal components by @ and also multiply them by the element total mass. 3. Set all off-diagonal components to zero. Applying this procedure to Eq. (8.1.21) results in 390 0 0 pAl}o PF 0 0 7/0 0 39 0 (8.1.28) oo00P This matrix is called the diagonal mass matrix and also conserves the mass for the translational degrees of freedom. Another technique to develop a diagonalized mass matrix is discussed in Refs [14,15] using numerical quadrature points located only at the nodes. ‘The clement stiffness matrix does not change for the dynamic analysis because the shape functions are the same for both static and dynamic analyses. However, the force term may vary as function of time. The force vector is for the dynamic analysis 1 {F*()} = i a(z,)[H]" dz (8.1.24) Thus, Eq. (8.1.24) is in general different from Eq. (8.1.12). As a result, the matrix equation for a dynamic beam analysis is after assembly of element matrices and vectors (Mi{a} + [K]{d} = {F@)} (8.1.25)242 Beam and Frame Structures Chapter 8 For free vibration of a beam, the eigenvalue problem is [16]. ({K] - w?[M]) {4} =0 (8.1.26) where w is the angular natural frequency in radians per second and {d} is the mode shape. Example problems for static, dynamic, and eigenvalue analyses of beams are provided at later sections using the MATLAB program. 8.2 Timoshenko Beam The Timoshenko beam theory includes the effect of transverse shear deformation. As a result, a plane normal to the beam axis before deformation does not remain normal to the beam axis any longer after deformation. Figure 8.2.1 shows the deformation in contrast to that in Fig. 8.1.2. While Galerkin’s method was used to derive the finite element matrix equation for the Euler-Bernoulli beam equation, the energy method is used for the present formulation for the Timoshenko beam. Let u and v be the axial and transverse displacements of a beam, respectively. Because of the transverse shear deformation, the slope of the beam @ is different from dv/dz. Instead, the slope equals (dv/dz) — y where is the transverse shear strain. As a result, the displacement field in the Timoshenko beam can be written as u(z, y) = —y0(z) (8.2.1) v(2)=0 (8.2.2) where the z-axis is located along the neutral axis of the beam and the beam is not subjected to an axial load such that the neutral axis does not have the axial strain. A beam subjected to both axial and transverse loads is considered in a next section for frame structures. From Eq. (8.2.1) and Eq. (8.2.2), the axial and shear strains are do feeias (8.2.3) dy yao4@ (624) As explained in the previous chapter on the energy method, the element stiffness matrix can be obtained from the strain energy expression for an element. The strain energy for an element of length / is b 1 ph/2 bp 1 ph/2 U= $f [ Ee dy de + of f{ yPGy dy de (8.2.5) 2So Jenje 2 Jo J-nj2 in which the first term is the bending strain energy and the second term is the shear strain energy. Moreover, b and h are the width and height of the beams respectively, and pis the correction factor for shear energy whose value is normally 2.Section 8.2 Timoshenko Beam 243 Yu Figure 8.2.1 Timoshenko Beam First, substituting Eq. (8.2.3) and Eq. (8.2.4) into Eq. (8.2.5) and taking integration with respect to y gives OND af dy\* dv U= A) (2) a(Z) dy dz + gf (-0+) ca(-0422) dy dz (8.2.6) where J and A are the moment of inertia and area of the beam cross-section. In order to derive the element stiffness matrix for the Timoshenko beam, the variables v and @ need to be interpolated within each element. As seen in Eq. (8.2.6), » and @ are independent variables. That is, we can interpolate them independently using proper shape functions. This results in satisfaction of inter-element compatibility, i.e. continuity of both the transverse displacement v and slope @ between two neighboring elements. As a result, any kind of C° shape functions can be used for the present beam element. Shape functions of order C* are much easier to construct than shape functions of order C1. It is especially very difficult to construct shape functions of order C? for two-dimensional and three-dimensional analyses such as the classical plate theory. ‘We use the simple linear shape functions for both variables. That is, v= [Hy Hy] { . } (8.2.7) 6 = [Hy Ho] { 5 } (8.2.8) where H; and Hy are linear shape functions. The linear element looks like that in Fig. 8.1.1, but the shape functions used are totally different from those for the Hermitian beam element, Using Eq. (8.2.7) and Eq. (8.2.8) along with the strain244 Beam and Frame Structures Chapter 8 energy expression Eq. (8.2.6) yields the following element stiffness matrix for the Timoshenko beam: [K*] = [Kj] + [Ky] (8.2.9) where 0000 EI}0 1 0 -1 KSl=T lo 0 0 0 (8.2.10) 0-10 1 4 2% -4 @ BGA | 2 aoa 4l |-4 -21 4 -2 2 ie [Ki]= (8.2.11) One thing to be noted here is that the bending stiffness term, Eq. (8.2.10), is obtained using the exact integration of the bending strain energy but the shear stiffness term, Eq. (8.2.11), is obtained using the reduced integration technique [17,18]. For the present calculation, the one-point Gauss quadrature rule is used as shown in the example given below. The major reason is if the beam thickness becomes so small compared to its length, the shear energy dominates over the bending energy. As seen in Eq. (8.2.10) and Eq. (8.2.11), the bending stiffness is proportional to ** while the transverse shear stiffness is proportional to hl, where h and | are the thickness and length of a beam element, respectively. Hence, as 4 becomes smaller for a very thin beam, the bending term becomes negligible compared to the shear term. This is not correct in the physical sense. As the beam becomes thinner, the bending strain energy is more significant than the shear energy. This phenomenon is called shear locking. In order to avoid the shear locking, the shear strain energy is under-integrated. Because of the under-integration the present beam stiffness matrix is rank deficient. That is, it contains some fictitious rigid body modes (i.e. zero energy modes). Example 8.2.1 shows the computation of the shear stiffness term. #& Example 8.2.1 We use the linear isoparametric element to integrate the shear energy term in Eq. (8.2.6) to produce the shear stiffness matrix Eq. (8.2.11). Using the concept of isoparametric mapping explained in Chapter 6, the shear stiffness term becomes -1/l 1—ry2]} fii nsce [-G [-3 (l+ny2 [KS The expression is a quadratic polynomial in terms of r so that the two-point Gauss quadrature will evaluate the integration exactly. For under-integration of one order less, we use the one-point Gauss quadrature rule. The integration point is 0 and the weight is 2. Applying this to Eq. (8.2.12) results in Eq. (8.2.11). ¢Section 8.3 Beam Elements with Only Displacement 245 ‘These kinds of beam elements can be derived for any order of shape functions higher than one. That is, beam elements can have three or larger number of nodes per clement depending on the order of shape functions. For each case, the shear stiffness matrix should be under-integrated consistently. The order of integration for the shear stiffness matrix is one less than what is required for exact integration. The consistent mass matrix for the Timoshenko beam is computed from 1 f[ PAIN] [Jaz (8.2.13) 0 where (8.2.14) 2010 ALI0 0 0 0 Ima PEt i720 (8.2.15) 0000 ‘The same lumped and diagonal mass matrices as given in Eq, (8.1.22) can be used for the present beam element. 8.3 Beam Elements with Only Displacement Degrees of Freedom In this section, we develop a family of beam elements which have only displace- ments as nodal degrees of freedom and no slope as nodal degrees of freedom [19]. In this aspect, these beam elements are similar to plane stress elements as given in Chap- ter 9. Therefore, when a beam needs to be discretized along its thickness direction as well as along its axial direction, these beam elements can be easily applied for the mesh. Figure 8.3.1 shows one example which uses stacked beam elements along the beam thickness. If there are multiple embedded cracks in a beam like interlaminar delamination in a laminated composite beam, it may require more beam elements through its thickness. If we plan to use beam elements which have displacements and rotations as nodal degrees of freedom for this application, we need special care at the interface of the neighboring top and bottom beam elements. As seen in Fig. 8.3.2, complicated constraint equations should be applied to maintain the continuous deformation across interface. Let us derive a linear beam element with displacement degrees of freedom only, which is the simplest element of this family. The element has six degrees of freedom which are axial and lateral displacements. There are axial displacements at the four corner points and lateral displacements at the two ends of the element as seen in Fig. 8.3.3. In Fig. 8.3.3 and the subsequent formulation, u represents the axial displace- ments at the corner points and v represents the lateral displacements at the ends.246 Beam and Frame Structures Chapter 8 Crack Poy fh Figure 8.3.1 A Beam with an Imbedded Crack (L >> h) The subscripts ‘1’ and ‘2’ refer to the left and right ends while the superscripts ‘” and ‘b’ indicate the top and bottom sides of the element, respectively. The displacement field of the element is ("Go } taney (3a) where [N] is the matrix of shape functions and {d°} is a vector of nodal displacements. The axial displacement is assumed to vary linearly along both axial and lateral directions. It can be written as 2 u(z, y) = > Ni(2) [Hi(y)uf + Ho(y)ul] (8.3.2) =I = Ni(z)Ha(y)uh + Ni(2)Ho(u)ul + No(2)Ha(y)uh + No(w)Ha(y)uy The lateral displacement, which is assumed to be constant through the thickness of the element, varies linearly along the axial direction and can be written as 2 oe) = OMe) 1 = Ni(z)vi + Na(z)v2 (8.3.3) Here, Nj and H; are the linear shape functions in the axial and lateral directions. The beam element may use a higher order shape function for the axial direction if there are more nodal points in the axial direction, and the linear shape function for lateral direction. However, in this study, the linear shape function is used for both N; and H; for simplicity. That is, (8.3.4)Section 8.3 Beam Elements with Only Displacement 247 J 4a Up Dome. Ma w= Us + eo Continuity at Interface hy on! ty 7 ert Mag= - dy 4 Ay ea Dos aD Pe Boe ~ BH Figure 8.3.2 Continuity Requirements at the Interface for Conventional Beam Elements ¥ “u Uy it t nL us b La | Ug x 0 U Figure 8.3.3 Four-Noded Beam Element with Six Degrees of Freedom where | and h are the length and height of the beam element, respectively, if the beam is a rectangular shape. For simplicity, notations, Nj, No, Hi and Hy will be used instead of Ni(z), N2(z), Hi(y), and Ho(y) in the following derivation. Axial normal strain can be written as du ON, ON, Ge one Ge a+ at ONa “Oz Hyud + ON? srt (8.3.5) t aus + Oz and the shear strain is du | av ay = Het (83.6)248 Beam and Frame Structures Chapter 8 _ OH, OHi | 0H; ON ON; | dy Nou, + oy Not + Gout ae oH: Mul + Beal + % The element stiffness matrix can be obtained by minimizing the total strain en- ergy which contains both bending and the transverse shear energy. This minimization yields the following element stiffness matrix {K*] = [K6] + [3] (8.3.7) where the subscripts ‘b’ and ‘s’ indicate bending and the transverse shear, respectively. ‘The bending and transverse shear stiffness matrices are given as tooph tsl= fl [10097 2 (B) dude (638) tooph tesla fi [18.97 6B.) aude (639) where E and G are the elastic and shear modulii of the beam and the vectors {Bs} and {B,} are derived below. The strain-displacement relationship for the axial strain and shear strain can be written as from Eqs (8.3.5) and (8.3.6) ée = {Bs} {d°} (8.3.10) and rey = {By} {a} (8.3.11) where {By} = {fn My, 0 en, Many 0 of =) (OH, — OH; ON, 0H; 0H, ON; {By} = {me Oy Nag: Gost Uae oy Na i my (8.3.12) me {d°} = {ub uf v1 ub uf v2} ‘The bending stiffness matrix can be obtained by carrying out the integration in Eq. (8.3.8) which will result in oo ecceReccueaeecceeaeccke 1 2 en 20 00 0 aa (K= alo -10 2 1.0 (8.3.18) pee kcceweeceteeccce: oe 0 0 0 0 0 0. For Eq. (8.3.9) the reduced integration technique is used along the z-axis to prevent shear locking which occurs when the ratio of beam length to beam thicknessSection 8.4 Mixed Beam Element 249 is large. That is, one-point Gauss quadrature is used for integration in the z-direction. This integration yields the transverse shear stiffness matrix of the form GP -GP Gh. GP GP — 2th -GP Ge -2Glh -GI? Ge 2GIh ee 2GIh =-2GlIh 4Gh? 2GlIh =-2GIh -4Gh? (K=gR| ce -ce acin ‘Ge -cr -2Gn| &44) -G? GP -2Gth -GP GP OGth -2GIh 2GIh —4Gh? —2GIh 2th 4h? The element stiffness matrix, which is obtained by adding the bending and transverse stiffness matrices, can be expressed in the following form a,+2a3 -a,;+a@3 a4 @;—2a3 ~a;—a3 —a4 =a; +43 a; +2a3 —aq —a—a3 a1 —2a3 ay Ik]=| % Ss = os (8.3.15) a, —2ay —a;—ay ag a +203 a, +03 —ay =a;—a3 a,—2d3 -a4 a, +03 a1 +203 ay a4 a4 a2 a4 a4 a2 where each symbol denotes ac ene ee ay (8.3.16) The mass matrix can be derived similarly as shown in previous sections. The lumped mass matrix for the linear element is 100000 010000 (00 20 00 M"l="Flo 00100 (83.17) C000 10 000002 Here pab is the element mass, and the element is assumed to have a unit width. Otherwise, the matrix is multiplied by the beam width. 8.4 Mixed Beam Element ‘This section develops mized beam elements {20,21]. The elements have the transverse deflection and bending moment as primary degrees of freedom. In the finite element method, the primary variables are more accurate than secondary variables which are usually obtained from derivatives of the primary variables, When the transverse deflection and slope are the primary variables, bending stress is a secondary variable which is related to derivative of the primary variables. On the other hand, the mized beam elements have the bending moment as a primary variable and the250 Beam and Frame Structures Chapter 8 bending stress is computed directly from the bending moment without taking any derivative. As a result, there is no loss in accuracy in computing the bending stress in the mized beam elements. The bending stress is usually one of the most important solutions needed in the beam analysis. In order to derive the mized elements, we consider the governing equations shown below: M_ @& BI a= (4. aM casa e (8.4.2) Galerkin’s method is applied to Eq. (8.4.1) and (8.4.2) using the same shape functions for both v and M. Then, for an element of length /, we obtain #H i ‘INITLN] de [M] + f : (ey [4] de [v] = [re]. (8.4.3) [ [z= my [= dz [M) = -f [N]fq dz + [rv (8.4.4) in which [NV] is the vector of shape functions, and @ and V are the slope and shear force, respectively. The element stiffness matrix from these equations becomes [KJ = FS a (8.4.5) where 1 [Ku] = a i [NFPIN] de (8.4.6) (Kia) = [Kai]? = [2 wy" [ae | 2 (8.4.7) For the linear mized beam element shown in Fig. 8.4.1, Eq. (8.4.5) becomes (K*\{a"} = {f°} (8.4.8) where 21 I 6EI —6EI 1 T 20 -6EI «6EI (l= Gani | enr -6er 0 0 (8-49) -6EI 6EI 0 0 {d°} = {Mi Mz v1 02)" (8.4.10) {f°} = {01 2 U-Q: Vi-Q:}” (8.4.11) and Q; is the equivalent pressure load applied to the nodal points.Section 8.4 Figure 8.4.1 Linear Mixed Beam Element Mixed Beam Element 251 Boundary conditions are applied to the element in the following way. For a simply supported node, both displacement v and bending moment M are set to zero while only displacement v and @ are set to zero at a clamped node. If a node is free without any applied moment, moment M is zero at the node, Example 8.4.1 shows the application of these boundary conditions. Any higher order shape functions may be introduced to Eq. (8.4.3) and Eq. (8.4.4) to obtain a stiffness matrix for a higher order mized beam element. # Example 8.4.1 Figure 8.4.2 shows a beam whose half is modeled using two linear mized beam elements. The left end is either simply supported or clamped and the right end is symmetric. Assembly of two beam elements gives the following system matrix equation. 2 1 0 G6EI -6EI 0 My, cn 1 4 1 ~6BI 12EI -6EI| | M2 0 1 0 1 2 0 -6BI 6EI||Ms|_| 0 6E1| 6£I -6EI 0 0 0 0 wu} [UW +05 -~6EI 12EI -6GEI 0 0 0 v2 1.0 0 -6EI 6EI 0 0 0 v3 0.5 (8.4.12) Since the right end is symmetric in this problem, the slope at the right end node (ie. the third node in Fig. 8.4.2) which corresponds to the third component in the right-hand-side column vector in this equation is set to zero, On the other hand, the slope at the left end (i.e. the first node) is given as 61 while the shear force at the left end support is given as Vj. Depending on the boundary condition, the slope may or may not be known. First of all, let us consider the simply supported left end. In this case, vy conditions to Eq. (8.4.12) gives 4 1 12EI —6EI ole 2 -6EI 6EI 6EI | 12EI -—6ET 0 0 -6EI 6EI 0 0 0 and M; = 0. However, 6) and Vi are unknowns. Applying these Mz 0 Ms 0 . a (8.4.13)252 Beam and Frame Structures Chapter 8 INfn Beam Rigidity =B7 node [ node 2 node 3 [m 1m 2m Figure 8.4.2 A Uniformly Loaded Beam Modeled Using Two Linear Mixed Beam Elements Solving this matrix equation provides v2 = —qBy and vg = —; for deflections, and also gives bending moments of 1.5 and 2 at nodes 2 and 3. Substituting these solutions back into Eq. (8.4.12) gives the shear force at the support equal to 2. These shear force and bending moments are exact solutions. Next, we consider the clamped left end. The corresponding boundary condition is v; = 0 and 6; = 0. Applying these conditions to Eq. (8.4.12) results in 2 1 0 ~6EI My, 0 1 1 4 1 12EI a M, 0 a 1 2 of ser Mz| =| 0 (8.4.14) -6EI 12EI —6EI v2 1.0 0 -6EI 6EI v3 05 The nodal deflections from this matrix equation are v2 = —g$7 and v3 = — 327. The bending moments are M; = —§, Mz = } and Mg = 3. The shear force at the left support is 2. In this case the shear force turns out to be exact but the bending moment is not exact. { The formulation provided above is based on the Euler-Bernoulli beam assump- tion. If the transverse shear deformation is included in the formulation, the governing equations are modified as given in Ref. [22} Pol eed (8.4.15) ot ag (8.4.2) where pt is the shear correction factor, h is the beam thickness and the beam is assumed to have unit width. Application of Galerkin’s method to these equations yields the element matrix shown as gey [Ku Ki vey = [fn Z (8.4.16)Section 8.5 Hybrid Beam Element 253 where [Ku] = a [orm + falas) dz (8.4.17) r (K12] = (Kail? = i : [¢ [= dz (8.4.18) ‘Again, the same shape functions are used for v and M in Eq. (8.4.17) and Eq. (8.4.18). For the linear element, Eq. (8.4.18) becomes +a P-a 6EI -6EI a P-a W2+a -6EI 6EI - ceeeeeeeeees (k 6EIl | 6£I —-6EI 0 0 (8.4.19) -6EI 6EI 0 0 where i °= Gh (8.4.20) We call the latter the thick beam element and the former the thin beam element because the effect of transverse shear deformation increases as the plate thickness to length ratio increases. Table 8.4.1 compares the thin beam and thick beam solutions for various loading and boundary conditions. The lumped mass matrix for the linear mized beam element is tut] = 2A z (8.4.21) sooo 0 0 00 10 01 coco 8.5 Hybrid Beam Element A hybrid beam element is introduced in this section. The hybrid element is based on the assumed strains within the beam element [23]. This element requires C® continuity. The formulation is based on a modified potential energy expression as given below for a beam with unit width 1 1 I= ii (-}4 Dees — FE Dies + FDilald} + EDsbald}) de — f[ {a}? {q} de 5 A (8.5.1) where : ae (8.5.2) a= -04% (8.5.3)254 Beam and Frame Structures Chapter 8 Table 8.4.1 Thin and Thick Beam Solutions for Various Beams with Unit Width ee Configuration Maximum deflections 3 P Thin Beam 1 Waa BEE g fl Thick Beam: PE, PL L ae ‘Wrae= SEI * 10GT : q : cs Y 4 Thin Beam :w,,= BET Thick Beam: L oh Tt Wma BET * OG a “ re Dt ee i ut CL “Waa 38481" BOGT 3 /: Thin Be =?e L/2 4 in Beam : w,> 788 7 Thick Beam: ee. pat Z Wat “4gEr * 4061 y 4 aut QUTTTTTy —|n Beem ae seat j}—________ Thick Beam: Sate L - aw a 7A-_—__—_________-} le Waa S641 BOGI P 3 y L/2 = PL Y Thin Beam :wi> ie2Er Thick Beam: 7 2 L PL PLY Se ‘Wrae= 19281 * 406K (E: Elastic modulus, G: Shear modulus, I: Moment of inertia of cross section)Section 8.5 Hybrid Beam Element. 255 {d} = {9 v}7 (8.5.4) and other parameters are defined below: Dy is the bending stiffness equal to EI, D, is the shear stiffness equal to uGA, Ly is the bending strain-displacement operator and L, is the shear strain-displacement operator. Invoking a stationary value of the equation results in the equilibrium equation and the generalized strain- displacement relation. In order to obtain the finite element model, generalized strains and displacements are discretized as the following: = [Ba]{as} (8.5.5) &s = [B]{as} (8.5.6) {a} = [N]{d} (8.5.7) where generalized strains are assumed independently within each element and gener- alized displacements are interpolated using generalized nodal displacement {d}. Thus, [By] and [B,] are matrices (or vectors for the beam problem) consisting of the poly- nomial terms of the generalized strain parameter vectors {a} and {a,}, respectively. Substituting Eq. (8.5.5) through Eq. (8.5.7) into Eq. (8.5.1) yields W=~ Has}TGilton} - Has}Geltas} + fou)" Hila} + fo} LH}{d} — {a}? {F} (8.5.8) where 1 [Gi] = f [Bi]? Ds[Bolde (8.5.9) 1 (G)= fe." DB, Jae (85.10) A 1 t= [teal Dstolm es (65.11) 1 t= [Ba D,clw lee (65.12) 0 and 1 {F}= f INT? (q}dz (8.5.13) 0 Invoking stationary values of Eq. (8.5.8) with respect to ay and a, respectively results in : —[Ge] {as} + [Ho}{a} = 0 (8.5.14) ~[Gslas} + [Hs]{d} = 0 (8.5.15) Eliminating {ay} and {a5} from Eq. (8.5.8), Eq (8.5.14) and Eq, (8.5.15) gives T= Stay? (aol Gal) + (HF Gs} -— {(T{F} (8.5.16)256 Beam and Frame Structures Chapter 8 Equation (8.5.16) finally gives the following finite element system of equations (K}{d} = {F} (8.5.17) in which (K] = [)" (Gi) el + (417 1G.) 1) (8.5.18) For a linear beam element, the generalized strain vectors are assumed as [B.) =[1 2] (8.5.19) and [B,) =1 (8.5.20) ‘These expressions represent that the bending strain varies linearly and the shear strain is constant within the linear beam element. Example 8.5.1 shows the derivation of the stiffness matrix for the linear beam element. The hybrid beam element can be also generalized for general higher order shape functions. & Example 8.5.1 Substituting the generalized strain vectors, Eq. (8.5.19) and Eq. (8.5.20), into Eq. (8.5.9) through (8.5.12) gives [@] = a [= | (8.5.21) [Ge] = wGAt (8.5.22) v= 3p at i] (8.5.23) [A = AL» 7 (8.5.24) Applying these expressions into Eq. (8.5.18) yields the same element stiffness matrices as in Eqs (8.2.9) through (8.2.11). The first term in Eq. (8.5.18) results in Eq. (8.2.10) while the second term in Eq. (8.5.18) yields that in Eq. (8.2.11). However, no reduced integration technique is used for the present stiffness matrix. } 8.6 Composite Beams Laminated composite beams are made of multiple layers which have in general different material properties. More general cases are dealt with in the chapter for plates and shells. In this section, we consider a simple case. The laminated beam is symmetric about the midplane axis so that there is no coupling between the inplane deformation and bending deformation. For this simple case, the beam formulations developed in the previous sections are directly applicable to the laminated beam. OneSection 8.6 Composite Beams 257 Figure 8.6.1 Laminated Beam thing to be generalized for the laminated beam is the beam rigidity. For a symmetric laminated composite beam with unit width, the equivalent beam rigidity is computed as (EDerie = 5 2 Ble? — af) (86.1) Here, n is the number of layers, and yi-1 and y; are the y—coordinate values of the bottom and top planes of the i** layer as seen in Fig. 8.6.1. In addition E; is the elastic modulus in the 2—direction of the i** layer. This equivalent beam rigidity is substituted into the previous beam elements to compute the stiffness matrices for laminated composite beams. However, one more important fact in laminated beam applications is the effect of transverse shear. Composite beams are not usually isotropic and their shear modulus is in general much lower than the elastic modulus. In this case, the shear deformation plays an important role [24,25]. For example, see Fig. 8.4.3. As the shear modulus G becomes much smaller than the elastic modulus in the thick beam solutions, the thick beam solutions deviate much more from thin beam solutions. As a result, thin beam solutions may not be accurate any more. In other words, the Euler-Bernoulli beam equation may not be suitable any more especially for rather thick laminated composite beams. As a result, the beam formulations including the effect of shear deformation can be used for analyses of laminated composite beams. In this case, the equivalent shear modulus is computed from (GA)egiv = D> bGi(yi — wi-1) (8.6.2) i=l where b is the width of the beam. The bending stress in a laminated composite beam can be determined from __MEBy (EDeriv (8.6.3) Oo258 Beam and Frame Structures Chapter 8 Figure 8.6.2 Laminated Beam Element where g; is the bending stress in the i** layer and y is a coordinate value in the i*® layer. Example 8.6.1 shows an application of beam elements with displacement degrees of freedom only developed in section 8.3 to an analysis of a laminated composite beam. # Example 8.6.1 Two different techniques can be used to model a laminated beam using beam elements with displacement degrees of freedom only. The first technique discretizes respective layers in the finite element analysis. As a result, the total number of elements is proportional to the number of layers in a laminated beam. The condensation technique can be applied to reduce the number of total degrees of freedom by eliminating internal layers degrees of freedom. This modeling technique is computationally expensive but it can describe a general shape of inplane deformation through the beam thickness. The second technique uses one beam element through the beam thickness regardless of the number of layers. This technique is computationally efficient but it assumes a linear deformation through the beam thickness. The development of this technique is described in the following paragraphs. Let uy and vp represent the axial and lateral displacements of the k* layer and let h be the beam thickness while hy and hy41 represent heights of the top and bottom sides of the k*" layer measured from the bottom of the beam (see Fig. 8.6.2). The relationship between the layer displacements and the global beam displacements can be written as ub, = eau} + cou} (8.6.4) ub, = coud + cau (8.6.5) ve =O (8.6.6) ub, = cud + cou} (8.6.7) ub a = esud + cauh (8.6.8) Ub,2 = 02 (8.6.9)Section 8.7 Two-Dimensional Frame Element 259 where constants c’s are h-h a= (8.6.10) h a=e (8.6.11) h-h =H (8.6.12) h ca = “HH (8.6.13) This relationship can be expressed in the matrix form ub cr e000 On pet uh 3400 0 01 [ul mil | 0 0 1 0 0 0) | wef = ]0 0 0 q e 0] Ju ee tha 0 0 0 cs co Of |us Oe 0000 01) by or in a short notation {a*) = (rita (86.18) where {d*} and {d} are the displacement vectors of the k*” layer and the beam, respectively. [7] is the transformation matrix shown in Eq. (8.6.14). Now, the stiffness and mass matrices of a laminated beam element can be expressed as be] = ere (8.6.15) feed: we) = Seren (8.6.16) & Here, [K*] and [M*] are the stiffness and mass matrices of the k** layer, respectively. Using this technique, a single beam element can include all the layers of a laminated beam. Figure 8.6.3 illustrates a simply supported laminated beam with four layers. The elastic modulus E} is assumed to be either 20 or 100 times greater than E2. The finite element solutions are provided in Fig. 8.6.4. Figure 8.6.5 also shows the solutions for the same beam but with eight layers. t 8.7 Two-Dimensional Frame Element A frame structure is made of many beam members connected together. It may be of planar or spatial geometry. For a planar frame structure, each beam member is generally subjected to both bending and axial loads as illustrated in Fig. 8.7.1. As a260 Beam and Frame Structures Chapter 8 Le 'P/2 P/e Figure 8.6.3 Simply Supported Laminated Beam with Four Layers *." EXACT solution *** FEM solution displacement, m. *.* EXACT solution *** FEM solution displacement, m. bo e+ ot 10 20 30 rn 80 @ length, m. Figure 8.6.4 Static Deflections of the Laminated Composite Beam with 4 Layers: E /E2=20, 100 for top and bottom result, a planar (2-D) frame element must include both axial and bending deformation.Section 8.7 Two-Dimensional Frame Element 261 *-* EXACT solution E001 + FEM solution : 52.09 3 3 20,09] 1. - 10 20 30 0 50 60 length, m. x10" *.* EXACT solution "** FEM solution 0 10 2 40 50 60 0 length, m. Figure 8.6.5 Static Deflections of the Laminated Composite Beam with 8 Layers: E,/E2=20, 100 for top and bottom If the deformation is small, we may neglect the coupling between the two deformations. As a result, the element stiffness matrix for a 2-D frame element can be constructed by superimposing both axial and bending stiffnesses. For example, the stiffness matrix of a linear 2-D frame element is using Hermitian beam element AP 0 0 -AP 0 0 0 lar GIL 0 -12I 6Il 0 el 41? 0 oll 2m? -AP 0 Oe Ale 0 0 0 -12r -6Il 0 121-611 0 Git ai 0 -6ll 4IP IK] = Zz (87.1) for the element degrees of freedom {u; v1 0; tp U2 42} as seen in Fig. 8.7.2. Other C° type beam elements may be used to develop a frame stiffness matrix. In general, a beam member is inclined to the global coordinate system as shown in Fig. 8.7.3. In this case, the element stiffness matrix requires the planar transformation. Figure 8.7.4 shows two coordinate systems: local and global systems.262 Beam and Frame Structures Chapter 8 P P PL PL B c B G — P A A Figure 8.7.1 A Planar Frame Structure with Free Body Diagrams ‘| | Y Uv, _==——— Figure 8.7.2 A Linear Frame Element The global system is denoted by a superimposed bar on both coordinate axes and displacements. The relation between the local and global displacements is uy coe 00 0 01 fa, n esc 0 0 0 01 |e 0 01 0 0 ol |e wf = |o0 00 ¢ s 0| ja Co) v2 000 -s ¢ 0| |e 0 02000 Oni ve, where ¢ = cos and s = sin. In a short notation, Eq. (8.7.2) can be written as {a°} = [THa} (8.7.3) Then, the stiffness matrix for a planar frame element is expressed in terms of the global coordinate system as given below: (Ke) = EP] (8.7.4) Carrying out the matrix multiplication gives a3 a4 Oe seeere ei ere cree a4 ag a7 a4 —ag a7 [kK] = | oF as Tar (8.7.5) —a3 —G4 —ds a3 a4 —ds5 aq —dg —a7 a4 ag az as a7 43 45 — a7Section 8.7 Two-Dimensional Frame Element YU VU 2 zu Zu Figure 8.7.4 Coordinate Transformation where ay = AEE ol 2EI a = ——— ao (A)2 + (BE)e 6EI w= (%2)c 263 (8.7.6) (8.7.7) (8.7.8) (8.7.9) (8.7.10) (8.7.11) (8.7.12)264 Beam and Frame Structures Chapter 8 8.8 Three-Dimensional Frame Element A beam member in a spatial (3-D) frame is generally subjected to axial, bending and torsional loads as illustrated in Fig. 8.8.1. If beam members have circular cross- sections, the element stiffness in a local axis is as seen in Fig. 8.8.2 Ki, Ke K« i Fi 8.8.1 w= [ih ae] where a, 0 0 0 0 0 0 bh 0 0 by Flo 0 ¢ 0 © oO (Knl=|9 9 — oO (8.8.2) ee eee ey 0 & 0 0 0 2%s -a, 0 0 0 0 0 0 -bh O 0 0 by eye 0 0 -a 0 -e 0 Hho 0 we oo (8.8.3) 0 0 co 0 cs (0 0 -b 0 0 0 b; 0 0 0 0 Oo of oo 0 —b -)./0 0 « 0 « 0 Kel=|9 9 0 a 0 0 (68.4) 0 0 @ 0 2% 0 0 -b, 0 0 0 2s eo (8.8.5) 12EI, 6EI, 2E1, ba, EE ya (88.8) a= Eh y= Se, y= ME (88.7) In these equations, Jy and I; are moments of inertia of the cross-section about y— and z—axes and J is the polar moment of inertia. The corresponding element degrees of freedom is {d*} = {ur v1 wi 0: 62 83 ur v2 tw 05 04 85)” (8.8.8) in which @; and 64 are the rotational degrees of freedom associated with the twisting moment, and 62, 43, 0s and 8¢ are slopes associated with bending moments. This stiffness matrix in terms of a local coordinate system needs to be transformed into that in terms of the global coordinate system in the same way as shown in Eq. (8.7.4). In this case, the transformation matrix [T] is of size 12 x 12.Section 8.9 MATLAB Application to Static Analysis 265 Figure 8.8.1 A Simply Supported Beam Figure 8.8.2. Spatial Frame Element 8.9 MATLAB Application to Static Analysis The static analysis of a beam or a frame is to solve the following matrix equation: [K]{q} = {F} (8.9.1) where the system stiffness matrix [K] and the system force vector {#} are obtained by assembling each element matrix and vector. Several m-files are written to compute an element stiffness matrix and a mass matrix, which is used for dynamic problems266 Beam and Frame Structures Chapter 8 100 0b clement 12 3 4 | gett {in node7m 2 3 4 5 6 i | ii : (| Cross section Be10X 10 PS! Figure 8.9.1 A Simply Supported Beam in the next section, for various beam and frame elements formulated in this chapter. The names of m-files are given below: febeami.m : Hermitian beam element (see Sec. 8.1) {febeam2.m : Timoshenko beam element (see Sec. 8.2) febeam§.m : beam element with displacement degrees of freedom (see Sec. 8.3) febeam4.m : mixed beam element (see Sec. 8.4) feframe2.m: 2D frame element (see Sec. 8.7) Detailed informations regarding these m-files are provided in Appendix A. The following examples show computer programs for finite element analyses of beam and frame structures written in MATLAB and the m-files described above. Example 8.9.1 Figure 8.9.1 shows a simply supported beam whose length is 20 in. The beam has also elastic modulus of 10 x 10° psi and its cross-section is 1 in. by 1 in. The beam is subjected to a center load of 100 Ib. We use 5 Hermitian beam elements for one half of the beam due to symmetry to find the deflection of the beam. Figure 8.9.1 shows the finite element discretization. The constraints applied to this problem are no deflection at the left boundary support (ie v1 = 0) and zero slope at the symmetric node (i.e. 05 = 0). Their system degrees of freedom are 1 and 12, respectively, in the present mesh. Furthermore, only a half of the center load is applied at the symmetric node because of symmetry. The finite element analysis program for the present static problem is listed below. % % BX891.m: MATLAB program to solve a static beam deflection using % Hermitian beam elements % % Variable descriptions % k = element stiffness matrixSection 8.9 MATLAB Application to Static Analysis % kk = system stiffness matrix % ff = system force vector % index = a vector containing system dofs associated with each element % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with the dofs in bedof % number of elements % number of nodes per element % number of dofs per node nnode=(nnel-1)*nel-+1; % total number of nodes in system ode*ndof; % total system dofs % elastic modulus % moment of inertia of cross-section % length of a half of the beam 0/nel; % element length of equal size % cross-sectional area of the beam % mass density (arbitrary value for this problem because % it is not used for the static problem) % option for mass matrix (arbitrary value and not used here) % first dof (deflection at left end) is constrained % whose described value is 0 bedof(2)=12; % 12th dof (slope at the symmetric end) is constrained % whose described value is 0 ff=2eros(sdof,1); % initialization of system force vector er0s(sdof,sdof); % initialization of system matrix =zeros(nnel*ndof,1); % initialization of index vector 50; % because a half of the load is applied due to symmetry nel % loop for the total number of elements =feeldofi(iel,nnel,ndof); _% extract system dofs for each element k=febeam1(el,xi,leng,area,tho,ipt);_% compute element stiffness matrix kk=feasmbl1 (kk,k,index); % assembly into system matrix end [kk,ff]=feaplyc2(kk,ff,bcdof,bcval); _-% apply the boundary conditions fool=kk\ff; % solve the matrix equation §-——_—________ % Analytical solution 1)*2; c=P/(48*e*xi); k=(i-1)*ndof+1; esol(k}=c*(3*1"2-4*x"2)*x; esol(k-+1)=c*(3#1°2-12*x"2); end %— % Print both exact and fem solutions 267Beam and Frame Structures Chapter 8 function {k,m]=febeam1(el,xi,leng,area,tho,ipt) % % Purpose: % Stiffness and mass matrices for Hermitian beam element % nodal dof v; theta: v2 thetay % % Synopsis: % [k,m]=febeam (el, xi,leng,area,tho,ipt) % % Variable Description: % k - element stiffness matrix (size of 4x4) % m - element mass matrix (size of 4x4) % el - elastic modulus % xi - second moment of inertia of cross-section % leng - element length % area - area of beam cross-section % tho - mass density (mass per unit volume) % ipt = 1: consistent mass matrix % — 2 lumped mass matrix % — otherwise: diagonal mass matrix ae % % stiffness matrix % c=el*xi/(leng3); k=c*[12 6*leng -12 6"leng;... 6*leng 4*leng? -6*leng 2*leng?; -12 -6*leng 12 -6*leng;... 6*leng 2*leng? -6*leng 4*leng3]; % % consistent mass matrix tho*area*leng/420; m*[156 22*leng 54 -13*len; 22*leng 4*leng? 13*leng -3*leng: 54 18"leng 156 -22*leng;... -13+leng -3*leng? -22+leng 4*leng3]; % % lumped mass matrix % elseif ipt==2Section 8.9 MATLAB Application to Static Analysis ol ros( 4,4); mass=rho*area*leng; ag({mass/2 0 mass/2 0)); m= % % diagonal mass matrix ros(4,4); mass=rho*area*leng; m=mass*diag( [1/2 leng2/78 1/2 leng3/78]); % end oe The finite element solution obtained from this MATLAB program as well as the exact solution are store = dof # femsol exact 1.0000 0.0000 0.0000 % deflection at node 1 2.0000 0.0030 0.0030 % slope at node 1 3.0000 0.0059 0.0059 % deflection at node 2 4.0000 0.0029 0.0029 % slope at node 2 5.0000 0.0114 0.0114 % deflection at node 3 6.0000 0.0025 0.0025 % slope at node 3 7.0000 0.0158 0.0158 % deflection at node 4 8.0000 0.0019 0.0019 % slope at node 4 9.0000 0.0189 0.0189 % deflection at node 5 10.000 0.0011 0.0011 % slope at node 5 11.000 0.0200 0.0200 % deflection at node 6 12.000 0.0000 0.0000 % slope at node 6 4 Example 8.9.2 We want to solve Example 8.9.1 using Timoshenko beam elements. The computer program is almost the same as that given in Example 8.9.1. Instead of calling febeam1.m we need to call febeam2.m to compute the element stiffness matrix. In the beginning of the program list shown in Example 8.9.1, we add sh=3.8*10°6; % shear modulus of the beam and the following line k=febeami (el,xi,leng,area,rho,ipt); is replaced by 269270 Beam and Frame Structures Chapter 8 k=febeam2(el,xi,leng,sh,area,tho,ipt); The computed solution is also compared to the exact answer below. store = dof# femsol exact 1.0000 0.0000 0.0000 % deflection at node 1 2.0000 0.0030 0.0030 % slope at node 1 3.0000 0.0059 0.0059 % deflection at node 2 4.0000 0.0029 0.0029 % slope at node 2 5.0000 0.0113 0.0114 % deflection at node 3 6.0000 0.0025 0.0025 % slope at node 3 7.0000 0.0158 0.0158 % deflection at node 4 8.0000 0.0019 0.0019 % slope at node 4 9.0000 0.0188 0.0189 % deflection at node 5 10.000 0.0011 0.0011 % slope at node 5 11.000 0.0200 0.0200 % deflection at node 6 12.000 0.0000 0.0000 % slope at node 6 4 Example 8.9.3 This example again solves the same problem in Example 8.9.1 using beam elements with displacement degrees of freedom. This beam clement is different from the beam elements used in previous examples. As a result, the complete program is included in the following. ‘% % EX893.m: MATLAB program to solve a static beam deflection using % beam elements with displacement degrees of freedom only % % Variable descriptions % k = element stiffness matrix % kk = system stiffness matrix % ff = system force vector % index = a vector containing system dofs associated with each element % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with the dofs in bedof % number of elements % number of nodes per element % number of dofs per node nel-1)*nel-+1; % total number of nodes in system sdof=nnode*ndof, % total system dofs % elastic modulus % shear modulus % length of a half of the beamSection 8.9 MATLAB Application to Static Analysis leng=10/nel; % element length of equal size : % height of the beam % width of the beam % mass density (arbitrary value for this problem because ‘% it is not used for the static problem) bedof(1)=3; % deflection at left end is constrained beval(1): % whose described value is 0 bedof(2)=16; % inplane displ. at the right end is constrained % whose described value is 0 bedof(3)=17; % inplane displ. at the right end is constrained beval(3)=0; % whose described value is 0 ff=zeros(sdof,1); % initialization of system force vector kk=zeros(sdof,sdof); % initialization of system matrix eros(nnel*ndof,1); % initialization of index vector % because a half of the load is applied due to symmetry % loop for the total number of elements index=feeldof1(iel,nnel,ndof); ‘% extract system dofs for each element k=febeam3(el,sh,leng,heig,width,tho); % compute element matrix kk=feasmbl1 (kk,k,index); % assembly into system matrix end [kk,ff]=feaplyc2(kk,ff,bedof,beval); -% apply the boundary conditions fsol=kk\ ff; % solve the matrix equation % % Analytical solution )*2; /(48*e*xi); k=(i-1)*ndof+1; esol(k-+2)=c¥(3*1°2-44x"2)*x; esol(k-+1)=c*(3*1"2-12*x"2)*(-0.5); esol(k)=c*(3*1°2-12*x"2)*(0.5); end ee % print both exact and fem solutions Inum? fsol esol’) function [k,m]=febeam3(el,sh,leng,heig, width,tho) %—__—_ % Purpose: % Stiffness and mass matrices for beam element with displacement % degrees of freedom only % nodal dof ue uf v; ub ub ve % 271Beam and Frame Structures Chapter 8 % Synopsis: % (km]=febeam | (el,sh,leng,heig,tho,area,ipt) % % Variable Description: % k - element stiffness matrix (size of 6x6) % m - element mass matrix (size of 6x6) % el - elastic modulus % sh - shear modulus % leng - element length % heig - clement thickness % width - width of the beam element % tho - mass density of the beam element (mass per unit volume) % lumped mass matrix only % % % stiffness matrix % al=(ch*leng* width)/(4*heig); a2=(sh*heig* width) /leng; a3=(el*heig* width) /(6*leng); a4=sh*width/2; k= [ a1+2*a3 -al+a3 a4 al-2*a3 -al-a3 -a4s... -al+a3 al+2*a3 -a4 -al-a3 al-2*a3 a4;... ad -ad a2 a4 -ad -a2;... al-2*a3 -al-a3 a4 al+2*a3 -al+a3 -a4;. -al-a3 al-2*a3 -ad -al+a3 al+2*a3 a -ad ad -a2 -ad ad a2]; % % lumped mass matrix % m=zeros(6,6); mass=rho*heig*width*leng/4; m=mass*diag([1 1211 2]); ann The solution output is store = dof # — femsol exact. 1,0000 0.0015 0.0015 % axial displ. at bottom side of node 1 2.0000 -0.0015 — -0.0015 % axial displ. at top side of node 1 3.0000 0.0000 0.0000 % transverse displ. at node 1 4.0000 0.0014 0.0014 % axial displ. at bottom side of node 2 5.0000 -0.0014 -0.0014 % axial displ. at top side of node 2 6.0000 0.0059 (0.0059 % transverse displ. at node 2 7.0000 0.0013 (0.0013 % axial displ. at bottom side of node 3 8.0000 -0.0013 -0.0013 % axial displ. at top side of node 3 9.0000 0.0113 0.0114 % transverse displ. at node 3 10.000 0.0010 0.0010 % axial displ. at bottom side of node 4Section 8.9 MATLAB Application to Static Analysis 11.000 -0.0010 += -0.0010 % axial displ. at top side of node 4 12.000 0.0158 0.0158 % transverse displ. at node 4 13.000 0.0005 0.0005, % axial displ. at bottom side of node 5 14.000 = -0.0005 = -0.0005 % axial disp|. at top side of node 5 15.000 0.0188 ~——0.0189 % transverse displ. at node 5 16.000 0.0000 0.0000 % axial displ. at bottom side of node 6 17,000 0.0000 0.0000 % axial displ. at top side of node 6 18.000 0.0199 0.0200 % transverse displ. at node 6 # Example 8.9.4 Solve the same example again using the mixed beam elements. The computer program list is provided below. % % EX804.m: MATLAB program to solve a static beam deflection % problem using mixed beam elements % % Variable descriptions % k = element stiffness matrix % kk = system stiffness matrix % ff = system force vector % index = a vector containing system dofs associated with each element % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in bedof %e nel=5; % number of elements % number of nodes per element ndof=2; % number of dofs per node nnode=(nnel-1)*nel-+1; % total number of nodes in system % total system dofs % bending moment at node 1 is constrained % whose described value is 0 % deflection at node 1 is constrained % whose described value is 0 ff=zeros(sdof,1); % initialization of system force vector kk=zeros(sdof,sdof); % initialization of system matrix index=zeros(nnel*ndof,1); % initialization of index vector % because a half of the load is applied due to symmetry % \oop for the total number of elements index=feeldofi(iel,nnelndof); _-% extract system dofs for each element k=febeam4(10°7,0.083333,2,0,1,1,1); % compute element stiffness matrix kk=feasmbl1 (kk, k,index); % assembly into system matrix end [kk, ff] =feaplyc2(kk, ff, bedof,beval); % apply the boundary conditionsBeam and Frame Structures Chapter 8 fsol=kk\ ff; % solve the matrix equation %$——_——_—__— % analytical solution for i = I:nnode 2; c=P/(48*e*xi); )*ndof+1; esol (k-+1)=c*(341°2-4*x"2)*x; esol(k)=-50*x; end %&—— % print both exact and fem solutions store=[num’ fsol esol’] ‘% — function [k,m]=febeamd(el,xi,leng,sh,heig,tho,ipt) %— % Purpose: % Stiffness and mass matrices for mixed beam element % bending moment and deflection as nodal degrees of freedom % nodal dof Mi v1 M2 v2 % % Synopsis: % [k,m]=febeam4(el,xi,leng,sh,heig,tho,ipt) % % Variable Description: % k - element stiffness matrix (size of 4x4) % m - element mass matrix (size of 4x4) % el - elastic modulus % xi - second moment of inertia of cross-section % \eng - length of the beam element % sh - shear modulus % heig - beam thickness % tho - mass density of the beam element (mass per unit volume) % ipt = 1 - lumped mass matrix % — = otherwise - diagonalized mass matrix % % % stiffness matrix % if sh =: % % thin beam (no shear deformation)Section 8.9 % MATLAB Application to Static Analysis I/leng 0 -1/leng 0;... leng/(6*el*xi) -1/leng leng/(3*el*xi) 1/leng;... -I/leng 0 1/leng 0 J; else % % thick beam (includes shear deformation) % a=6/(5*sh*leng*heig); k= [ 1/(3*el*xi) +a 1/leng 1/(6*el*xi) Tfleng 0 -1/leng 0;.. 1/(6*el*xi)-a -1/leng 1/(3*el*xi)-+a 1/leng;... -I/leng 0 i/leng 0 J; % end % % lumped mass matrix % if ip % m=zreros(4,4); mass=rho*heig*leng/2; m=diag([0 1 0 1); % diagonal mass matrix % else % m=zeros(4,4); mass=rhotheig*leng/2; m=mass*diag({1 1 1 1)); % end % [leng/(3*el*xi) 1 /leng leng/(6*el*xi) -1 /leng;... -1/leng;... The solution from the computer program is given below. store = dof # 1.0000 2.0000 3.0000 4.0000 5.0000 6.0000 7.0000 8.0000 fem sol 0.0000 0.0000 -100.00 0.0059 200.00 0.0114 -300.00 0.0158 exact 0.0000 0.0000 100.00 0.0059 200.00 0.0114 -300.00 0.0158 % bending moment at node 1 % deflection at node 1 % bending moment at node 2 % deflection at node 2 % bending moment at node 3 % deflection at node 3 % bending moment at node 4 % deflection at node 4 275276 Beam and Frame Structures Chapter 8 y 20 in 2 @ le» 1 in Cross section 60 in E=10X10° psi ( 4 elements at 15 in each) Figure 8.9.2 An L-Shaped Frame 9.0000 -400.00 -400.00 % bending moment at node 5 10.000 0.0189 0.0189 % deflection at node 5 11.000 -500.00— -500.00 % bending moment at node 6 12.000 0.0200 0.0200 % deflection at node 6 4 Example 8.9.5 Find the deflection of a frame of L-shape (see Fig. 8.9.2) which is made of two beams of lengths of 60 in. and 20 in., respectively. Both beams have cross-sections of 2 in. height by 1 in. width, The elastic modulus is 30 x 10° psi. The frame is subjected to a concentrated load of 60 Ib at the end of the smaller beam and one end of the long member is fixed. Use 6 elements to find the deflection of the frame. The MATLAB program is written below using 2-D frame elements. % % EX895.m: MATLAB program to solve static deflection for a 2-D frame % % Variable descriptions % x and y = global x and y coordinates of each node % k = element stiffness matrix % kk = system stiffness matrix % ff = system force vector % index = a vector containing system dofs associated with each element % bcdof = a vector containing dofs associated with boundary conditions % beval = a. vector containing boundary condition values associated with % the dofs in bedof %Section 8.9 MATLAB Application to Static Analysis % number of elements % number of nodes per element % number of dofs per node nnode=(nnel-1)*nel+1; % total number of nodes in system sdof=nnode*ndof, % total system dofs x(1)=0; y(1): % x, y coords, of node 1 in terms of the global axis % x, y coords. of node 2 in terms of the global axis % x, y coords. of node 3 in terms of the global axis % x, y coords. of node 4 in terms of the global axis % x, y coords. of node 5 in terms of the global axis % x, y coords. of node 6 in terms of the global axis % x, y coords. of node 7 in terms of the global axis % elastic modulus % cross-sectional area % moment of inertia of cross-section % mass density per volume (dummy value for static analysis) % transverse deflection at node 1 is constrained % whose described value is 0 % axial displacement at node 1 is constrained % whose described value is 0 % slope at node 1 is constrained % whose described value is 0 ff=zeros(sdof,1); % initialization of system force vector r08(sdof,sdof); % initialization of system matrix index=zeros(nnel*ndof,1); % initialization of index vector % load applied at node 7 in the negative y direction for iel=1: % \oop for the total number of elements index=feeldofl (iel,nnel,ndof); % extract system dofs for each element nodel=iel; % starting node number for element “ie!” node2=iel+1; % ending node number for element ‘ie!’ nodel); v1: ‘nodel); % x and y coordinate values of ’nodel’ x2=x(node2); y2=y(node2); % x and y coordinate values of ’node2” leng=sqrt((x2-x1)*2+(y2-y1)°2); % length of element ‘ie!’ if (x2-x1 % compute the angle between the local and global axes beta=pi/2; else beta=atan((y2-y1)/(x2-x1)); end k=feframe2(cl,xi,leng,area,tho,beta,1); % compute element matrix kk=feasmbl1 (kk,k,index); % assembly into system matrix end {kk,f{]=feaplyc2(kk,ff,bedof,beval); _-% apply the boundary conditions fsol=kk\ ff; % solve the matrix equation and print % ae % Print both exact and fem solutions store=[num’ fsol]Beam and Frame Structures Chapter 8 % function [k,m]=feframe?2(el,xi,leng,area,rho, beta ipt) % % Purpose: % Stiffness and mass matrices for the 2-D frame element % nodal dof u; vi theta; uz v2 thetag % % Synopsis: % (k,m}=feframe2(el,xi,leng,area,tho,beta,ipt) % % Variable Description: % k - element stiffness matrix (size of 6x6) % m - element mass matrix (size of 6x6) % el - elastic modulus % xi - second moment of inertia of cross-section % leng - element length % area - area of beam cross-section % rho - mass density (mass per unit volume) % beta - angle between the local and global axes % is positive if the local axis is in the ccw direction from % — the global axis ‘% ipt = 1 - consistent mass matrix % = 2- lumped mass matrix % = 3-~ diagonal mass matrix % % % stiffness matrix at the local axis % a=el*area/leng; 0 12%c 6*leng*c 0 -12* c 6*leng*c;... 0 G*leng*c 4*leng*2*c 0 -6*leng*c 2*leng“2*c;... -200200;., 0 -12%c -6*leng*c 0 12*c -6*leng*c;... 0 Gtleng*c 2*leng”2*c 0 -6*leng*c 4*leng”2*c]; % % rotation matrix % t=[ cos(beta) sin(beta) 000 -sin(beta) cos(beta) 0 0 0 0}... 0010060;... 0.0 0 cos(beta) sin(beta) 0;. 0.0 0 -sin(beta) cos(beta) 03... 000001); % % stiffness matrix at the global axisSection 8.9 MATLAB Application to Static Analysis % kar'*kl*y; % consistent mass matrix % if ip' % mm=rtho*area*leng/420; ma=tho*area*leng/6; ml=[2*ma 0 0 ma 0 03... 0 156*mm 22*leng*mm 0 54*mm -13*leng*mm;... 0 224leng*mm 4*leng"2*mm 0 13*leng*mm -3*leng”2*mm;... ma 0 0 2*ma 0 0;... 0 54*mm 13*leng*mm 0 156*mm -22*leng*mm;... 0 -13*leng*mm -3*leng”2*mm 0 -22*leng*mm 4*leng”2*mm); % % lumped mass matrix a elseif ipt==2 % ml=zeros(6,6); mass=tho*area*leng; ml=mass*diag({[0.5 0.5 0 0.5 0.5 0)); % % diagonal mass matrix % else % ml=zeros(6,6); mass=rtho*area*leng; ml=mass*diag({0.5 0.5 leng”2/78 0.5 0.5 leng”2/78]); end % % mass in the global system % m=r*ml*r, %— The finite clement solution is compared to the exact solution at some selected nodes as given below: store = dof # — femsol —_ exact 1.0000 0.0000 0.0000 % horizontal displ. at node 1 2.0000 0.0000 0.0000 % vertical displ. at node 1 3.0000 0.0000 0.0000 % slope at node 1 4.0000 0.0068 % horizontal displ. at node 2 5.0000 0.0000 % vertical displ. at node 2 6.0000 —-0.0009 % slope at node 2 279280 Beam and Frame Structures Chapter 8 7.0000 0.0270 % horizontal displ. at node 3 8.0000 0.0000 % vertical displ. at node 3 13.000 0.1080 0.1080 % horizontal displ. at node 5 14.000 -0.0001 % vertical displ. at node 5 15.000 -0.0036 —-0.0036 % slope at node 5 16.000 0.1080 % horizontal displ. at node 6 17.000 -0.0386 % vertical displ. at node 6 18.000 0.0040 % slope at node 6 19.000 0.1080 0.1080 % horizontal displ. at node 7 20.000 -0.0801 0.0801 % vertical displ. node 7 21.000 -0.0042 -0.0042 % slope at node 7 8.10 MATLAB Application to Eigenvalue Analysis Eigenvalue problems of a beam or a frame structure are solved using the finite element method written in MATLAB programs. The m-files described in the previous section compute both the element stiffness and mass matrices so that they are used in the present programs in order to compute the natural frequencies of a beam or a frame structure. To this end, we need to assemble element stiffness and mass matrices into the system stiffness and mass matrices. One m-file used here is feaplycs.m : application of constraints to both mass and stiffness matrices This m-file modifies the eigenvalue matrix equation with given constraints. Instead of redimensioning the matrix size because of the constraints, the original matrix size is conserved. However, the modified eigenvalue matrix equation will contain fictitious zero eigenvalues in the same number of the constraints. As a result, the user should exclude these zero eigenvalues from the computer solution. Except for these, the structure of computer programs is the same as that in examples in the last section. The following examples show the computer programs written in MATLAB to compute the natural frequencies. 4 Example 8.10.1 Find the natural frequencies of a free beam of unit Jength. It has a cross-section 1 by 1 and it has also mass density of 1. The elastic modulus of the beam is 12. All the units are consistent. Use 4 elements to model the whole beam so that nonsymmetric mode shapes can be included. Use Hermitian beam elements and consistent mass matrices. The computer program is listed below: %- % EX8101.m: MATLAB program to find the natural frequencies of a freeSection 8.10 MATLAB Application to Eigenvalue Analysis % beam using Hermitian elements % % Variable descriptions % k = element stiffness matrix % m = element mass matrix % kk = system stiffness matrix % mm = system mass matrix % index = a vector containing system dofs associated with each element % bcdof = a vector containing dofs associated with boundary conditions vector containing boundary condition values associated with the dofs in bedof % number of elements % number of nodes per element % number of dofs per node nnode=(nnel-1)*nel-+1; % total number of nodes in system sdof=nnode*ndof; % total system dots % elastic modulus % moment of inertia of cross-section % mass density % total length of the beam Jeng=tleng/nel; % uniform mesh (equal size of elements) area=1; % cross-sectional area i : % flag for consistent mass matrix eros(sdof,sdof); % initialization of system stiffness matrix mm=zeros(sdof,sdof); % initialization of system mass matrix index=zeros(nnel*ndof,1); % initialization of index vector for iel=1:nel % loop for the total number of elements index=feeldofl(iel,nnel,ndof); _% extract system dofs for each element [k,m]=febeam 1 (el,xi,leng,area,rho,ipt); % compute element matrices kk=feasmbl1 (kk,k,index); % assembly of system stiffness matrix mm=feasmbl1(mm,m,index); % assembly of system mass matrix end fsol=eig(kk,mm); % solve the eigenvalue problem fsol=sqrt(fsol) % print circular frequencies A— $e The finite element solution is compared to the exact solution below: mode # femsol = exact 0 0.0000 0.0000 % rigid body mode a 0.0000 0.0000 % rigid body mode 2 22.400 22.373 % first non-zero circular frequency 3 62.060 61.673 % second non-zero circular frequency 4 121.86 120.90 % third non-zero circular frequency 5 223.29 178.27 % fourth non-zero circular frequency ‘As seen in the comparison, the two solutions agree well for lower frequencies. However, the discrepancy becomes larger for higher natural frequencies. In order282 Beam and Frame Structures Chapter 8 to obtain more accurate higher modes, the finite element model should have a refined mesh to represent the corresponding mode shapes properly. As a result, if we want to improve the fourth non-zero frequency in this example, we need to refine the mesh. t Example 8.10.2 Find the natural frequencies of a cantilever beam whose length is 1 m long. The beam has the cross-section of 0.02 m by 0.02 m and the mass density is 1000 Kg/m®. The elastic and shear moduli are 100 GPa and 40 GPa, respectively. Use 4 elements. The MATLAB program is shown below. % % EX8102.m: MATLAB program to solve the natural frequencies of % a beam using beam elements with displacement degrees of % freedom only % % Variable descriptions ‘% k = element stiffness matrix % kk = system stiffness matrix % m = element mass matrix % mm = system mass matrix % index = a vector containing system dofs associated with each element % bedot vector containing dofs associated with boundary conditions % number of elements % number of nodes per element ndof=3; % number of dofs per node nnode=(nnel-1)*nel-+1; % total number of nodes in system sdof=nnode*ndof; % total system dofs el=100*10°9; % elastic modulus % shear modulus % total beam length % same size of beam elements % height (or thickness) of the beam % width of the beam % mass density of the beam % bottom inplane displ. at node 1 is constrained bedof(2): % top inplane displ. at node 1 is constrained bedof(3)=3; % transverse displ. at node 1 is constrained kk=zeros(sdof,sdof); % initialization of system stiffness matrix mm=zeros(sdof,sdof); % initialization of system mass matrix index=zeros(nnel*ndof,1); % initialization of index vector for iel=1:nel % loop for the total number of elements index=feeldofl(iel,nnel,ndof); _% extract system dofs for each element [k,m]=febeam3(el,sh,leng,heig,width,tho); _% compute element matrices kk=feasmbli (kk,k,index); % assembly of system stiffness matrix mm=feasmbl1(mm,m,index); % assembly of system mass matrixSection 8.10 MATLAB Application to Eigenvalue Analysis end [kk,mm]=feaplycs(kk,mm,bedof); % apply the boundary conditions fsol=eig(Ikk,mm); % solve the matrix equation and print fsol=sqrt(fsol) %— The natural frequencies obtained from the finite element program are compared to the exact solutions mode # fem sol —_exact 1 200.00 203.00 % first circular natural frequency 2 1260.0 1272.0 % second circular natural frequency 3 4040.0 3562.0 % third circular natural frequency Example 8.10.3 Find the natural frequencies of a frame of L-shape which is made of two beams of length of 1 m each as seen in Fig. 8.9.2. Both beams have cross-sections of 0.01 m by 0.01 m. The elastic modulus is 100 GPa. The beam has mass density of 1000 Kg/m*. Use 10 elements. % % EX8103.m: MATLAB program to find the natural frequencies for a 2-D % frame using frame elements % % Variable descriptions % x and y = global x and y coordinates of each node % k = element stiffness matrix % kk = system stiffness matrix % m = clement mass matrix % mm = system mass matrix % index = a vector containing system dofs associated with each element % bedof = a vector containing dofs associated with boundary conditions % number of elements % number of nodes per clement % number of dofs per node nel-1)*nel+1; % total number of nodes in system node*ndof; % total system dofs x(1)=0; y(1 % x, y coord. of node 1 in terms of the global axis % x, y coord. of node 2 in terms of the global axis % x, y coord. of node 3 in terms of the global axis % x, y coord. of node 4 in terms of the global axis % x, y coord. of node § in terms of the global axis % x, y coord. of node 6 in terms of the global axis % x, y coord. of node 7 in terms of the global axis % x, y coord. of node 8 in terms of the global axis % x, y coord. of node 9 in terms of the global axis 283284 Beam and Frame Structures el=100*10°9; Chapter 8 ; % x,y coord. of node 10 in terms of the global axis % x, y coord. of node 11 in terms of the global axis % clastic modulus area=0.0001; % cross-sectional area xi: 8.3333*10°(-10); % moment of inertia of cross-section % mass density per volume % transverse deflection at node 1 is constrained % axial displacement at node 1 is constrained % slope at node 1 is constrained % initialization of system stiffness matrix % initialization of system mass matrix eros(nnel*ndof,1); % initialization of index vector ‘1:nel % loop for the total number of elements index=fecldofl(iel,nnelndof); _% extract system dofs for each element % starting node number for element ’iel’ % ending node number for element ’iel’ % x and y coordinate values of ’nodel’ % x and y coordinate values of ’node2’ leng=sqrt((x2-x1)"2+(y2-y1)°2); % length of element ‘ie!’ if (x2-x1)==0; — % compute the angle between the local and global axes beta=pi/2; else beta=atan((y2-y1)/(x2-x1)); end [k,m]=feframe2(el,xi,leng,area,tho,beta,1); % element matrix kk=feasmbl1 (kk,k,index); % assembly of system stiffness matrix mm=feasmbl1(mm,m,index); % assembly of system mass matrix end [kk,mm]=feaplycs(kk,mm,bedof); % apply the boundary conditions fsol=eig(kk,mm); % solve the matrix equation and print fsol=sqrt(fsol) % ‘The numerical solutions are mode # fem sol. 1 34 % first circular natural frequency 2 92 % second circular natural frequency 3 455 % third circular natural frequency 4 667 % fourth circular natural frequency 8.11 MATLAB Application to Transient Analysis In the transient analysis of a structure, the equation of motion at time t is [MI{d}‘ + [C}{dy + Ka}! = {FY (8.11.1)Section 8.11 MATLAB Application to Transient Analysis 285 where [M], [C] and [K] are the system mass, damping, and stiffness matrices and they are assumed to be independent of time. Superscript t denotes time. We will present the direct time integration scheme to solve Eq. (8.11.1). There are many integration techniques which can be applied to the matrix equation, Readers may refer to Refs [16-18]. In this section, the central difference scheme is explained because it is one of the most popular techniques in the structural mechanics application. ‘There are two versions of the central difference scheme. The first method is summarized below. Detailed derivation of this technique is provided in (16]. 1. Compute system matrices like [M], [C], and [K]. 2. Solve for the initial acceleration {d}° from {a}° = [M}"? ({F}° — [C]{a}° — [K]{4}°) where {d}° and {d}° are the initial displacement and velocity vector. 3. Compute the fictitious displacement at time At from {d}~4* = {d}° — (As){dy? + 4 (a)? 4, Compute the effective mass matrix. [M1] = ail] + zaq(C] 5. Repeat 6 through 9 for each time step. 6. Compute the effective force vector. {F}t = {F}' — ((K] ~ dalM)){u}! — (gM) — sag [C]) {a} 7. Find the displacement at time t + At from {d}ttat ale [Fy 8. Find the acceleration at time t. {d}' = gia({a}tt — 2{d} + {a} 44) 9. Find the velocity at time t. {a} = ha(taytt4t ~ {ayers where At is the time step size. ‘The second form of the central difference scheme, called summed form [19], is described below. 1. Repeat 2 through 4 for each time step. 2. Compute the acceleration. {d}t = [MI-({F} — [C]{d}* — [K]{4}*) 3. Compute the velocity from the acceleration. {dyt+0-5at — {dyt-0.50 4 At(dyt 4. Compute the displacement from the velocity. {d}itOt = {a}t 4 At{d}tto-sat ‘The both central difference techniques are conditionally stable. The critical time step size for stability is Aterst = Fain (8.11.2) where Tmin is the smallest period of the discretized system with finite degrees of freedom. Therefore, the time step size At must be smaller than or equal to this critical size to maintain numerical stability. Comparison of the two central difference techniques are discussed below. (1) ‘The first technique computes nodal displacements, velocities, and accelerations at the same time steps while the second technique computes them at different time286 Beam and Frame Structures Chapter 8 steps. That is, nodal displacements and accelerations are determined at the same time steps but velocities are found at the middle of the time steps. As a result, in order to compute both kinetic and strain energies, for example, at the same time step, an interpolation technique is used for the second central difference scheme to have both displacements and velocities at the same time steps. (2) The first technique does not have to calculate nodal velocities and accelerations to march along the time if a user does not need them. On the other hand, the second technique needs to computes all of them to progress along the time. (3) In terms of computer programming, the second method is much easier than the first method. (4) Both techniques require initial solutions at some fictitious time steps. The first scheme needs the displacements at time —At while the second requires the velocities at time —0.5A¢ to initiate the computations. The first technique has the consistent way to determine the solution at the fictitious time step but the second technique does not have that. Hence, the first technique can be used to find the velocity solution at the fictitious time step for the second technique and the procedure for the second scheme is used after that. In other words, we compute {d}~“ from the first central difference technique. Then the fictitious velocity vector {d}-°*4* is obtained from jy-o.sas _ {a}? — {dh {dyosat = SO (8.11.3) Numerical experimentation was conducted in Ref. [30]. Both central difference schemes were applied to a crack propagation problem. The study showed that solutions from the both schemes were almost identical. In the study, the fictitious velocity for the second technique was assumed to be the same as the initial velocity at time 0. The solution obtained with this assumption and the second method was almost identical to the solution from the first method. The following example shows the second central difference scheme applied to a beam problem. # Example 8.11.1 Find the transient response of a cantilever beam whose length is 1 m long. The beam has the cross-section of 0.02 m by 0.02 m and the mass density is 1000 Kg/m*. The elastic modulus is 100 GPa. The beam is initially at rest and subjected to a constant tip load of 100 N at time 0. ‘The following computer program uses the second central difference method to determine the transient response. Four Hermitian beam elements are used. The critical time step size for this finite clement system is 1.149 x 10~* sec. The program uses At = 1 x 1074 sec. The program is listed below. ee % EX8111.m: MATLAB program to find the transient response % of a cantilever beam with a tip load % % Variable descriptions % k = element stiffness matrix % kk = system stiffness matrixSection 8.11 MATLAB Application to Transient Analysis % m = clement mass matrix % mm = system mass matrix % index vector containing system dofs associated with each element % bedof = a vector containing dofs associated with boundary conditions % acc = acceleration of nodal variables % vel = velocity of nodal variables % disp = displacement of nodal variables %- % number of elements % number of nodes per element % number of dofs per node nnode=(nnel- 1)*nel-+1; % total number of nodes in system sdof=nnode*ndof; % total system dofs el=100*10°9; % elastic modulus tleng=1; % total beam length leng=tleng/nel; % same size of beam elements xi=0.02°4/12; % height (or thickness) of the beam area=0.004; % cross-sectional area of the beam % mass density of the beam ipt=1; % option flag for mass matrix (consistent mass matrix) dt=0.0001; % time step size i % initial time % final time % number of time steps % number of constraints % transverse displ. at node 1 is constrained % slope at node 1 is constrained kk=zeros(sdof sdof); % initialization of system stiffness matrix mm=xeros(sdof,sdof); % initialization of system mass matrix force=zeros(sdof,1); % initialization of force vector index=zeros(nnel*ndof,1); % initialization of index vector acc=zeros(sdof,nt); % initialization of acceleration matrix 08 (sdof,nt); % initialization of velocity matrix -zeros(sdof,nt); ‘% initialization of displ. matrix vel(:,1)=zeros(sdof,1); % initial zero velocity disp(:,1)=zeros(sdof,1); % initial zero displacement force(9)=100; % tip load of 100 for iel=1:nel % loop for the total number of elements index=feeldof1 (iel,nnel,ndof); % extract system dofs for each element [k,m]=febeam1(el,xi,leng,area,tho,ipt); % compute element matrices kk=feasmbl1(kk,k,index); % assembly of system stiffness matrix mm-=feasmbl1(mm,m,index); % assembly of system mass matrix end mminy=inv(mm); % invert the mass matrix % central difference scheme for time integration mt % time integration loop it)=mminy* (force-kk*disp(;,it)); % compute acceleration % application of constrained conditions 287288 Beam and Frame Structures Chapter 8 for i=1:nbe % loop for number of constraints ibe=bedof(i); % nodal dof where constraint is applied acc(ibe,it)=0; % acceleration at the constrained dof set to 0 end +1)=vel(:,it) +acc(:,it)*dt; % compute velocity disp (:,it+1)=disp(;,it)+vel(:,it +1) *dt; % compute displacement end ace(:,nt-+1)=mminv*(force-kk*disp(:,nt+1)); % accel. at last time step %— Se % plot of the tip deflection plot(time,disp(9,;:)) xlabel("Time(seconds)’) ylabel(’Tip displ. (m)’) %— The plot of the tip deflection is shown in Fig. 8.11.1. The tip deflection shows an oscillation around the static deflection. Figure 8.11.1 also shows the same deflection when the time step size is 1.15 x 1074. Because the time step size is larger than the critical step size, the deflection becomes unstable and diverges. t 8.12 MATLAB Application to Modal Analysis of Undamped System In this section, the dynamic analysis of multiple degrees of freedom systems is presented and applied to beam structure examples. The multiple degrees of freedom systems are quite different from single degrees of freedom systems in terms of mathematical formulations and associated time responses. For multiple degrees of freedom system, we define modes which represent each component of overall dynamic responses. The modes are essential in describing the nature of motion and provide physical understanding of the dynamic behavior of the system. The modes are characterized by so-called eigenvalues and eigenvectors of the system. The eigenvaules are related to usually natural frequencies and eigenvectors to mode shapes of the given system. Unfortunately, these eigenvalues and eigenvectors are limited mostly to linear systems. This limitation is not significant, in fact, the majority of the dynamic systems are represented by linear systems. There are also plenty of computer software tools available to linear system analyses. Some key concepts are introduced in this section including a solution technique for cigenvalue/eigenvector problems, which is sometimes called the modal analysis. ‘The time response of a system is obtained in a straightfoward manner once the modal analysis results are ready. We consider an undamped system here and a damped system is discussed in the subsequent section.Section 8.12 Modal Analysis of Undamped System 289 Te displ. (m) Rooe ae e 04 ‘Time(seconds) Tip dispt. (m) of 02 ‘Time(seconds) Figure 8.11.1 Time Responses using Central Difference Method For a given n degree of freedom linear second order system, the governing differential equation of motion is described by the second order matrix equation as (MI{d} + (K] {d} = {F} (8.12.1) We seek to find the natural motion of the system, i.e. response without any forcing function. The form of response or solution is assumed as {d(t)} = {a} e** (8.12.2) where {4} is the mode shape (eigenvector) and w is the natural frequency of the motion. In other words, the motion is assumed to be purely sinusoidal due to zero damping in the system. The general solution turns out to be a linear combination of each mode as {d(i)} = erro" + co{da}eet + ---+ en{dn ent (8.12.3) where each constant(c;) is evaluated from initial conditions. Substituting Eq. (8.12.2) into Eq. (8.12.1) with {F} = 0 yields [31,32] (-w?[M] + [K]){d}el" = 0 (8.12.4) The above equation has a nontrivial solution if (—w?[M] + [K]) becomes singular. In other words, there exist n number of ws which satisfy |= 4?[M) + [AJ] = | — AM] + [KJ] =0 (8.125)290 Beam and Frame Structures Chapter 8 where A = w? is the eigenvalue of the system. Equation (8.12.5) produces solutions, w},w3,...,w2. Since the mass matrix is positive definite and the stiffness matrix is at least positive semidefinite, all w;s are nonnegative. This can be easily proven from w[M]{ $i} = (K] {60} (8.12.6) for the i eigenvalue and eigenvector. Let us multiply {4:}7 on both sides of the equation oP {bs} IMO.) = {60} KOs} (8.12.7) According to the general property of mass and stiffness matrix in the form {x}"[M]{x} > 0, {x}"[K]{x} >0 for {x} #0 In addition, the eigenvectors are orthogonal to each other, which can be easily shown from wP[M]{ds} = [KV b} (8.12.8) a3 [M]{4j} = [K]{ 65} (8.12.9) For proof, we premultiply {¢;}7 on both sides of Eq. (8.12.8) and subtract the transpose of Eq. (8.12.9) which is post-multiplied by {di}. The result becomes (w} — 9} ){O5} IM Gi} = (45) LANG} — (45) [A] {06} = 0 (8.12.10) Therefore, if i # j {6;}(MI{4s} = {45}? [KI{i} = 0 (8.12.11) and the eigenvectors are orthogonal with respect to the mass and stiffness matrices. The above orthogonality property includes systems with non-repeated rigid body degrees of freedom. For multiple rigid body modes, for example a three dimensional translational motion, a special form of orthogonality exists (see Ref. [16] for the special form). Orthogonality of eigenvectors in conjunction with positive and positive semidefiniteness of mass and stiffness matrices of a vibrational system is one of the distinct features of linear dynamic systems. The orthogonality of eigenvectors provides a useful normalization technique in the form {bi} IMHO} =1, {45} [KM{0s} = 0? (8.12.12) Once the eigenvectors are normalized, the following coordinate transformation is proposed {d} = (®] {n} (8.12.13) where [2] = [61, $2, ---s bal (8.12.14) is called modal matriz whose columns consist of normalized eigenvectors, and {n} = [n1,72)-.-)m]” is the vector of modal coordinates. Substitution of Eq. (8.12.14) into Eq. (8.12.1) yields (MI[O]{n) + [X18] {n} = (F} (8.12.15)Section 8.12 Modal Analysis of Undamped System 291 Next, we premultipy &7 on both sides of Eq. (8.12.15), so that [8]? Min} + [8] [KIA] {n} = 07 (F} (8.12.16) According to the orthogonality in Eq. (8.12.12), Eq. (8.12.16) can be rewritten as {n} + diaglu?] {n} = 87 {F} = {fi} (8.12.17) In other words, the system of equations are decoupled. tie tw Pn = fi (8.12.18) where i = 1,2,...,n, and f; is the i** row of 67{F}. Equation (8.12.18) represents the modal coordinate form of equations of motion, for which each independent vibrational mode is described by a decoupled second order differential equation. The modal coordinate equations are so useful since they provide the analytical solution for each mode. Also, the input function into the i** modal coordinate (f;) represents how much the mode is excited from the external input. 4 Example 8.12.1 Consider a Euler-Bernoulli beam model with one end fixed asin Fig. 8.12.1. For simplicity, the beam is modeled by two finite elements using the consistent mass matrix. The numerical data for the structure are p (linear mass density)=0.024 kg/m, EJ= 6.09 N-m?, and D=1.27 m. AMATLAB m-file called femodal.m produces the following mass and stiffness matrices after applying the boundary condition 0.0929 0 0.0161 -0.0967 0 1.4881 0.0967 ~0.5580 (I= 0161 0.0967 0.0464 ~0.1637 —0.0967 -0.5580 -—0.1637 0.7440 and 0.0052 0 —0.0026 0.0326 [kK] = 104 x 0 1.0880 -0.0326 0.2720 —0.0026 -0.0326 0.0026 —-0.0326 0.0326 0.2720 0.0326 0.5440 ‘The corresponding degrees of freedom for these matrices are {v1 01 v2 62} as shown in Fig. 8.12.1. The natural frequency and modal matrix are computed from the MATLAB function file femodal.m as follows 1 3.6692 1.3594 ~2.9157 —0.4570 1.9102 we | _ } 23.1786 [8] = —0.0931 0.0351 0.6871 0.7852 w3{ ) 78.3943 [* ~ | 4.0089 4.0304 4.4925 7.5441 w4 227.5337 —0.1102 0.3890 -0.8665 2.9165292 Beam and Frame Structures Chapter 8 < S il 4 SSS Figure 8.12.1 Two Elements Beam Model and the modal input force matrix [®]" {F} = [-4.0039 4.0394 — 4.4995 7.5441)" where the original input influence matrix is given by {F}=[0 01 of” ‘The MATLAB source file is provided below as a reference function [Omega,Phi, ModF]=femodal(M,K,F); %o % Purpose % ‘The function subroutine femodal.m calculates modal parameters h for a given structural system. It calculates natural frequency and % eigenvector. The eigenvectors are normalized so that the modal % mass matrix becomes an identity matrix. % % Synopsis: % — (Omega, Phi, ModF]=femodal(M,K,F) % % Variable Description: % Input parameters - h M, K - Mass and stiffness matrices % F - Input or forcing function % Output parameters - % Omega - Natural frequency(rad/sec) in ascending order % Phi - Modal matrix with each column corresponding to % the eigenvector. % ModF - Modal input matrices. ———— disp(’ ’) disp(’Please wait! - The job is being performed.’)Section 8.12 Modal Analysis of Undamped System 293 % Btseacsiaeeaetecnte ates eee EEEESE % Solve the eigenvalue problem and normalized the eigenvectors O———————————————_—_————————— [n,n]=size(M);[n,m]=size(F); [V,D]=cig(K,M); [lambda,k]=sort(diag(D)); V=V(:k)s Factor=diag(V*M*V); Phi=V*inv(sqrt(diag(Factor))); Omega=diag(sqrt(Vnorm’*K*Vnorm)); Modf=Vnorm”*F; %e Note that each modal coordinate or finite element degree of freedom can be taken as the output variables of femodal.m t In order to find out the solution to Eq. (8.12.18), the Laplace transformation technique is used. (0) + (0) ’ nafs) = HO + HO), fils) aot | ar (8.12.19) where 7;(0) and %(0) are related to the initial conditions as explained below. Taking the inverse Laplace transform of Eq. (8.12.19) yields the time domain solution ) U3 wy ; mult) = ni(O)cosuit + 9 sinwit + i A sinui(t —r)fi(r\dr (8.12.20) oi ‘As one might have expected, the solution consists of two parts: i) excitation by initial condition and ii) response due to the external forcing input, The convolution integral in the solution is not easy to evaluate in general, except for some special cases such as impulse and step inputs. ‘As shown above the initial conditions (7(0), 1(0)) for the modal coordinate are needed for the complete solution. This information can be directly obtained from the original transformation equation, Eq. (8.12.13). That is, {d(0)} = [8] {n(0)} (8.12.21) so that {n(0)} = [©]? (Mf) {4(0)} (8.12.22) Similarly {4(0)} = [6 Lanta) {410} (8:12.23) Now the solution of each modal coordinate is combined together to produce the solution in physical coordinates. {d(t)} = [8] {n()} (8.12.24)294 Beam and Frame Structures Chapter 8 ee ee i 1, 025 0.1 02 03 04 06 O7 0.8 og 1 05 Time(sec) Response at the tip a Vertical displacement ¢ $6 = 05 06 07 08 08 1 Time(sec) Figure 8.12.2 Impulse Responses by Modal Analysis In other words, a(t) = S> bing (t), for i=1,2,...,n (8.12.25) j=l # Example 8.12.2 _ In this example, the same model is used as in Example 8.12.1 to demonstrate the evaluation of impulse response of the beam. The impulsive force is applied at the tip of the beam. The analytical solution, Eq. (8.12.20), is incorporated into a MATLAB m-file feiresp.m. The initial conditions, {d(0)} and {d(0)} are set to zero. ‘The time response results are presented in Fig. 8.12.2. Note that the response time interval is so critical to show higher modes in the response. If the time interval is too large, higher modes will not show up in the response. This issue will be discussed later in frequency response analysis in Sec. 8.14. The source MATLAB m-file is presented below: function [eta,yim]=feiresp(M,K,F,u,t,C,q0,dq0); Ps % Purpose: % — The function subroutine feiresp.m calculates impulse response % — for a given structural system using modal analysis. It uses modal % — coordinate equations to evaluate modal responses anaytically, then % convert modal coordinates into physical responses %Section 8.12 Modal Analysis of Undamped System 295 Synopsis: [eta, yim]=feiresp(M,K,F,u,t,C,q0,dq0) Variable Description: Input parameters - M, K - Mass and stiffness matrices F - Input or forcing function u- Index for excitation + - Time period of evaluation C- Output matrix 0, dq0 - Initial conditions Output parameters - eta - modal coordinate response yim - physical coordinate response RRM Me ae a ST ak a a disp(’”) disp(’Please wait ee % Solve the eigenvalue problem and normalized the eigenvectors ee [n.n]=size(M);[n,m]=size(F); ine (”); J=cig(K,M); flambda,k]=sort(diag(D)); V=V(k); ag(V*M*V); V*inv (sqrt(diag(Factor))); omega=diag(sqrt(Vnorm’*K*Vnorm)); Fnorm=Vnorm’*F; oo % Find out impulse response of each modal coordinate analy tically _————— ee norm”*M%q0; detad: n phase=omega(i)*t; ta0(i)*cos(phase’)-+deta0(i)*sin(phase’)/omega(i)+... sin(phase’)*Fnorm(i,u); end % Convert into physical coordinates yim=C*Vnorm*eta’; % - The job is being performed:”) ‘norm’*M*dq0; eta=zeros(nstep,n);296 Beam and Frame Structures Chapter 8 8.13 MATLAB Application to Modal Analysis of Damped System For an degree of freedom system with inherent damping, the governing equation of motion can be written as (Md) + (1d) + [KI{d} = {F) (8.13.1) where [C] is a n by n damping matrix. The above system is stable due to the introduction of the damping term as explained below. The damping can be classified into inherent structural damping or damping by active control. The stability of the above system can be discussed by taking the total energy(kinetic plus potential) of the system U = FAT ned) + Hae (8.13.2) Assuming free vibration with {F}=0, the time rate of change of U becomes oF = tanta + 1K) (6.13.3) furthermore, using Eq. (8.13.1) @ = ay (0Ktd} (8.13.4) Therefore, as long as the damping matrix [C] satisfies {d}[Cl{d} > 0, for {d} #0 (8.13.5) it follows & <0 (8.13.6) and the system is stable with respect to the equilibrium state ({d},{d}) = (0,0). Estimating the damping matrix for a physical system is not easy in general. ‘There are some methods of modeling the damping matrix. One of the special cases is to use so-called proportional damping or Rayleigh damping in the form [C] = o[M] + 61K] (8.13.7) where a and f are constants. In other words, the damping matrix is proportional to the mass and/or stiffness matrix. The proportional damping has an advantage of possessing the same characteristic as the mass or stiffness matrix. The eigenvectors obtained from the mass and stiffness matrices conserve orthogonality with respect to the damping matrix. That is {JIC} =0, for 145 (8.13.8) and []7 [C][®] becomes a diagonal matrix. Now, the original governing equation, Eq. (8.13.1), can be rewritten in the modal coordinate form ik + awit twin = fi (8.13.9)Section 8.13 Modal Analysis of Damped System 297 Application of the Laplace transform yields mi(O) + (8 + 2s mi(0) + Fils) (8) = (62 + wis +0?) (8.13.10) ‘The inverse Laplace transform of n;(s) becomes & nit) =ni(O)e"S**coswat + (f(0) — ni(0))e~%"** sin(wat) i + + eS) ginwa(t — 7) f(r)dr (8.13.11) a Jo where wa = wi/1— G is the damped natural frequency and w; is the undamped natural frequency. In most practical cases, the modal damping ratio ¢; is less than unity so that the damped natural frequency is smaller than the undamped natural frequency. The modal coordinate solution in Eq.(8.13.11) can be used to produce the solution of the physical coordinates {d(t)} = [®1{n(@)} (8.13.12) where [6] is the modal matrix obtained from the damped system. 4 Example 8.13.1 In this example, we test the impulse response of a damped system. The system damping matrix is assumed as a proportional damping, and the mass and stiffness matrices are the same as in Example 8.12.1. The proportional constants are chosen as @=0.2, =0.005. The same unit impulsive force at the tip, as in Example 8.12.1, is applied. A MATLAB fediresp.m file is written as presented below. The simulation results are presented in Fig. 8.13.1. They show the damped responses at the two different positions of the beam. function [eta,yim]=fediresp(M,K,F,u,t,C,q0,dq0,a,b); xz % Purpose: % — The function subroutine fediresp.m calculates impulse response % for a damped structural system using modal analysis. It uses modal % coordinate equations to evaluate modal responses anaytically, then % convert modal coordinates into physical responses Synopsis: [eta, yim]=fediresp(M,K,F,u,t,C,q0,dq0,a,b) RRL ee Variable Description: % Input parameters : M, K - Mass and stiffness matrices F - Input or forcing influence matrix u- Index for excitation RR298 Beam and Frame Structures Chapter 8 x10" Middle response 2 - — i 8 Wer} 3 | ee ae 1012141818 20 Timetsoe) Ar Tip response Vertical displacement oy 2 rs of * | 5 12 14 16 18 20 10 Time(se0) Figure 8.13.1 Impulse Responses for a Damped System % t - Time of evaluation % u - Index for the excitation % C - Output matrix % 40, dq0 - Initial conditions % a, b- Parameters for proportional damping [C]=a[M]+b[K] % Output parameters : eta - modal coordinate response % yim - physical coordinate response % disp(’ ’) disp(’Please wait!! - The job is being performed.’) 6— % Solve the eigenvalue problem and normalized the eigenvectors %— ee = {n,n]=size(M);[n,m]=size(F); nstep=size(t’)}; [V,D}=eig(K,M); [lambda,k]=sort(diag(D)); % Sort the cigenvaules and eigenvectors =V(:,k); Factor=diag(V*M*V); Vnorm=V*inv(sqrt(diag(Factor))); % Eigenvectors are normalized omega=diag(sqrt(Vnorm’*K*Vnorm)); % Natural frequencies Fnorm=Vnorm’*F; ——————————— % Compute modal damping matrix from the proportional damping matrix % Modamp=Vnorm’*(a*M-+b*K)*Vnorm;Section 8.14 Frequency Response Analysis 299 zeta=diag((1/2)*Modamp*inv(diag(omega))) if (max(zeta) >= 1), disp(’Warning - Maximum damping ratio is greater than or equal to 1’) disp(’You have to reselect a and b ’) pause disp(‘If you want to continue, type return key’) end eta0=Vnorm’*M*q0; % Initial conditions for modal coordinates deta0=Vnorm’*M*dq0; % - both displacement and velocity eros(nstep,n); mm % Responses are obtained for n modes -omega(i)*sqrt(1-zeta(i)"2); phase=omegad*t; tcons=zeta(i)*omega(i)*t; eta(:,i)=exp(-tcons)’.*(eta0(i)*(cos(phase’)+-zeta(i)/sqrt(1-zeta(i)"2)*... sin(phase’))-+deta0 (i)*sin(phase’) /omegad+sin(phase’)*Fnorm(i,u)... /omegad); end % ee % Convert modal coordinate responses to physical coordinate responses ———————— yim=C*Vnorm*eta’; % 8.14 MATLAB Application to Frequency Response Analysis ‘The previous modal analysis of a system is mainly based upon the time domain approach. ‘The eigenvalues and eigenvectors directly produce solutions in time domain in the form of time response functions. Modal coordinates make it possible to derive sets of decoupled equations of motion. Each individual modal coordinate solution is combined to result in the physical coordinate solution. The modal analysis provides very convenient tools for understanding behavior of multiple degrees of dynamic systems. Sometimes, the time domain analysis is not the best choice, especially for modal testing and other applications. One supplementary approach is the frequency domain analysis. The frequency domain method has major advantages over the counterpart, ie., time domain analysis. In fact, it is being more widely adopted in signal processing, active control system design, modal testing, etc. Most of the vibrational systems can be characterized by their inherent frequency components which dictate both time and frequency responses. One key advantage of the frequency domain analysis is that one can span a whole range of frequencies which is not possible or impractical in the time domain analysis. Conversion of time domain signals into frequency domain signals and vice versa is relatively easy due300 Beam and Frame Structures Chapter 8 to the modern computational power. In this section, discussion on the frequency domain analysis is presented. The Fast Fourier Transform (FFT) and evaluation of the frequency response function for multiple degrees of freedom systems are presented. Consider a general continuous time domain signal given by z(t). The signal z(t) can be periodic or nonperiodic. It can be represented in the following expression 2-2 i X(M)ea0 (8.14.1) 2s where X(Q) is the Fourier transform of the time signal z(t) c2 X(Q) = i x(t)e** de (8.14.2) ‘The time signal z(t) is also called the inverse Fourier transform of X(Q). For the existence of the Fourier transform the following condition should be satisfied for x(t): [ * Ie(d)ldz (8.14.3) should have a finite value. The above constraint is not strict in the sense that it covers a wide range of signals of actual dynamic systems. On the other hand, introducing a new variable f = 9/2n, we have 2(t) = i X(feledg (8.14.4) and X(f)= [ (the FOr dt (8.14.5) Since the Fourier transform involves integral of general time varying complex vari- ables, it is not easy to carry out the integration. Except for some special cases, a numerical integration technique is needed. One efficient algorithm is to use the Dis- crete Fourier transform. The numerical integration is conducted by a finite number of summation at discrete points. Assume that there are N sampled values as m=x(t), te=kAt, &=0,1,2,...,.N-1 (8.14.6) Based upon the sample data points, we assume that the time domain data project into the corresponding frequency domain data. In other words, the Fourier transform is defined for the N discrete frequency points. (fy fas. fn) (8.14.7) where the frequency points should be in the range of the so-called Nyquist critical frequency feat (8.14.8)Section 8.14 Frequency Response Analysis 301 In other words, the sampling period (At) should be at least a half of the period of a signal to sufficiently represent the signal. oo N-1 X(fn) = [ a(t olde ~ J ape at ar (8.14.9) k=0 The above equation is called the Discrete Fourier transform. The Discrete Fourier transform (DFT) has the symmetry property with respect to the input frequency (fa). In other words, n X(fn)=X(fr-n), fa = War n=— vz N ye (8.14.10) and only a half of the transform is needed to represent all frequency components. The maximum frequency rage is given by 1 0 Hi(2,y)us (9.1.16) i=l 3 o(2,u) = Yo Hie, ude (9.1.17)310 Elasticity Problem Chapter 9 ‘These displacements can be also expressed as ul_[Hh 0 He 0 Hs 0]}wl_ {t}= [ 0m 0 mM 0 a = [N]{a} (9.1.18) where {d} ={u: 0 uz v2 ug 03)" is the nodal displacement vector. Use of this expression for strains yields ou Oh 9 a 9 Og 3 [oh 9 ek on pa] 0 Be 0 ho al ty (9.1.19) auf bv OH, Sif, 0H, OF, pH, Olle dy + br dy Os tye By We use symbol [B] to denote the matrix expression in the above equation. That is, ou oe {j= { eB \ = [B\{d} (9.1.20) ‘oy + be Galerkin’s method states w) = Hi (i = 1,2,3) and w, = H; (i = 1,2,3). Applying these weighting functions and Eq. (9.1.20) into Eq. (9.1.15) gives for the finite element domain integral farina (9.1.21) iE in which Q¢ denotes the element domain. As a result, the element stiffness matrix for elasticity can be expressed as [kK] = i (ei payaa (9.1.22) Equation (9.1.22) holds for any kind of element in any dimension. Evaluation of the linear shape function provides 1 [ (2-4) 0 (ys— 1%) 0 (vi = v2) 0 Bl=a (223) 0 (21 ~ 23) 0 (22-21) (t3— 22) (Ye—ys) (21-23) (ys—m) (t2-21) (t —¥2) (9.1.23) Substitution of Eq.(9.1.23) into Eq. (9.1.22) results in [Kk] = f. i [B]" [D][B\do = [By] [D][B\A (9.1.24)Section 9.2 Force Vector 3ll since both [B] and {D] are constant matrices independent of z and y. Here A is the area of the element. This expression is true for both plane stress and plane strain conditions. The material property matrix [D] is selected properly for plane stress (i.e. Eq. (9.1.6)) and plane strain (i.e. Eq. (9.1.7)) conditions. We assume a unit thickness for the plane stress condition because the solution is independent of the thickness direction for this case. However, if we want to include the thickness, the matrix in Eq. (9.1.24) is multiplied by the thickness. When other kinds of shape functions are used for the plane stress/strain condition, we just need to develop matrix [B] as shown in Eq. (9.1.20) and put it into Eq. (9.1.22). The size of row of [B] is always three for the plane stress/strain condition while the size of column equals twice the number of nodes per element because there are two degrees of freedoms per node. 9.2 Force Vector The two right-hand-side terms in Eq. (9.1.15) are the force vector. The first term is due to body forces and the other is due to tractions. The body force term is a domain integral. As a result, the same computation can be performed to this term as the stiffness matrix. Applying Galerkin’s method to this term in an element domain yields ine foi }o= fle 2l{g}eo=Lor{z}e ean where [WV] is defined in Eq. (9.1.18). On the other hand, the traction vector is a boundary integral. This boundary integral is very similar to what is described in Sec. 5.4. Let us consider a traction as shown in Fig. 9.2.1. The traction term can be evaluated as given below: Cee ae where sm and 8 ate the coordinate values along the temporary boundary axis s, and m and n are the two nodes on the element boundary where the traction is described. If the traction is constant, the traction vector becomes Um sq@cos0 tm | _ 1) sabsino tn { ~ 2) sabcos0 (9.2.3) Um sabsind in which 84 = 8m — Sm is the length of the boundary segment and @ is a constant traction value.312 Elasticity Problem Chapter 9 n ge oO ™m =Gcosv Pa? Ss Gy =Fsind s —— — . Sn td + Sm Sm Figure 9.2.1 Boundary Traction 4 Example 9.2.1 Find the nodal displacements in Fig. 9.2.2. We use two linear triangular elements as seen in the figure. Each element is also shown in a separate figure indicating global node numbers and local node numbers necessary for construction of the element stiffness matrix. Using the plane stress condition, the element stiffness matrix for the first element is given below along with the associated nodal displacements: 73° 3.3 -5.3 -20 -20 -137 (uy 3.3 7.38 13 -2.0 -2.0 -5.3] ] » 53 -13 53 00 00 13 us -20 -20 00 20 20 00 03, 20 -20 00 20 20 00 U2 -13 -5.3 13 00 00 53 v2 {K"]{d'} = 10° (9.2.4) ‘The stiffness matrix of the second clement is the same as that given in Eq, (9.2.4) because the two elements have the same size and shape as well as the same way of local node numbering. Assembling the elements and computing the nodal forces from the given traction using Eq. (9.2.3) results in [K]{d} = 73° 33 -2.0 -13 -53 —2.0 00 0.0 UL 3.3 7.3 -20 -5.3 -13 -2.0 00 0.0 v1 2.0 -20 73 00 00 33 -53 -13 uz 10° -13 -53 00 73 33 00 -20 -20 ae -5.3 -13 00 33 73 00 -20 -20 us —20 -20 33 00 00 73 -13 -5.3 v3 0.0 00 -53 -2.0 -20 -13 73 33 U4 0.0 00 -13 -20 -20 -53 33 7.3 04Section 9.3 Energy Method 313 : $= 10° N/m —— — } 2m { 3 Vo 2m. +=" 20 2® 4 Q + local node number global node number F=10 Cpa v 20,25 unit thickness 10 3@ 3@ Figure 9.2.2 Square Plate With Tangential Traction = (9.2.5) where Fir, Fiy, Fae and Fay are unknown forces while the essential boundary conditions state uj=v1=u3=v3=0. Applying these conditions to Eq. (9.2.5) and solving for the unknown nodal displacements give u2=6.135 x 1077, v9=1.450 x 1077, u4=6.975 x 1077 and v4=-1.660 x 1077. This is a very crude mesh so that these solutions are not accurate. { 9.3 Energy Method The total potential energy denoted by II consists of two parts: internal energy U and external energy W which is equal to work done by external loads. That is, T=U-w (9.3.1)314 Elasticity Problem Chapter 9 The internal energy is the strain energy caused by deformation of the body and can be written as = ; f fo}? {an (9.3.2) in where {c} = {or oy Tey}? denotes the stress and {ce} = {€ €y yey}7 is the strain. Equation (9.3.2) also holds for three-dimensional state of stresses. Use of the constitutive equation for Eq. (9.3.2) gives | ; i {9 (Wigan (9.3.3) since [D] = (DJ. On the other hand, the external work can be written as we ftw oH{f bane ftw on{ Se bar (9.3.4) Substitution of Eqs (9.3.3) and (9.3.4) into Eq. (9.3.1) and discretization of the domain into a number of finite element domains yields a=>om (9.3.5) oat i [ie le m=} f corwnaan—f tw oy {han [ te ofS bar @s0) For each finite element, applying Eqs (9.1.18) and (9.1.20) to Eq. (9.3.6) gives m= Hay? [tor orwiantay (a? [wr {4} an ~ {at f NP {% } ar (9.3.2) In order to find the equilibrium solution, we apply the principle of minimum total potential energy. The principle states: Of all kinematically admissible configurations, the deformation producing the mini- mum total potential energy is the stable equlibrium condition, Invoking the stationary value for Eqs (9.3.5) and (9.3.7) using this energy principle, we obtain ome eS —= [K*]{d} — {F} — {}) =0 9.3.8 Lama Xt Hd} — {F} - {8}) (9.3.8) where [K‘]= LE (B}? (/DI[Byan (9.3.9)Section 9.4 Three-Dimensional Solid 315 (y= [we ie } ao (9.3.10) {8} = [wr {bar (9.3.11) Here, Eqs (9.3.9), (9.3.10) and (9.3.11) are the element stiffness matrix, body force vector and surface traction vector, respectively. By comparing Eq. (9.1.22) to Eq. (9.3.9), it can be shown that Galerkin’s method results in the same matrix equation as the energy method. In addition, the force terms are identical to those obtained from Galerkin’s method. Especially, Eq. (9.3.11) looks different from Eq. (9.2.2) but they are the same when actual calculation is performed. 9.4 Three-Dimensional Solid The governing equations for three-dimensional elasticity are given below. Equations of equilibrium: Oe, Arey Ores et ay t +fe=0 (9.4.1) Orey , dey , OT ay Sy ae = toe (9.4.2) Dror | Oe oe +f =0 (9.4.3) Oz ay * where stresses are shown in Fig. 9.4.1 in the positive direction and fy, fy and f, are body forces per unit volume. Constitutive equation for an isotropic material: {o} = [Dl{e} (9.4.4) where fo}= {os oy os Tey Ty: Tex)” (9.4.5) {e}={te fy & tev Ws Yes)” (9.4.6) 1 yov 0 0 0 Poly 0 0 | = E y eo 0) 0 0 Borie) 0 F 0 ee GG ee 0 0 o 6 ee oO 0 0 0 0 0 Te316 Elasticity Problem Chapter 9 2 OG; Figure 9.4.1 Three Dimensional State of Stress Kinematic equation for small displacements: (9.4.8) reese + + + oats seeker skelesle where u, v and w are displacements in the z, y and z directions, respectively. Traction boundary condition: B, = 0gNz + TeyNy + Tet; = Ge (9.4.9) By = TeyNz + OyNy + Ty2Mz = By (9.4.10) ®, = Tez + Tysty +O, =O, (9.4.11) where n,, ny and n, are cosine directions of the outward unit normal vector on the traction surface and $ is the known value. ‘We want to derive the element stiffness matrix for a tetrahedron element as seen in Fig. 9.4.2. The element has four nodes. The shape functions for this element can be derived as given below: Let us assume a linear function in terms of z, y and z. ay az ob =x} (9.4.12) a4 u=[l ¢ y 2]Section 9.4 Three-Dimensional Solid Gung : (pg %3) pups) (Cap Uyr 2) z Figure 9.4.2 Tetrahedron Element Evaluation of u at every node yields {u} = [X]{A} where lan a 1 22 yo ze 1 23 ys 23 1 24 vs % = 317 (9.4.18) (9.4.14) Inverse of matrix [X] in Eq. (9.4.13) and substitution of the resulting expression into Eq. (9.4.12) gives u= (XIX) {u} = [A ]{u} where the shape functions are (H)=(Ai(e,y,z) He(2,y,z) Ha(z,y,z) Ha(z,y,2)] We use the same shape functions for the three displacements. = (I{d} (9.4.18) (9.4.16) (9.4.17)318 Elasticity Problem Chapter 9 where {d} is the nodal displacement vector. Substituting Eq. (9.4.17) into the three dimensional kinematic equation Eq. (9.4.8) results in fe} = [B]{4} (9.4.18) in which m o 0 #% o 0 % 0 0 4 o 0 0 Yo Go MH 0 0 Hh 0 Oo MO 0 0 0 0 4 0 0 3 oO 0 oH [B]=] om om G om of 9 Hs Os G Oh Oe ay ar ‘Oy or oy or ‘Oy oz 0 8h oh 9 Ha 9H, Q OH, OH, Q SHy OMe oz by Oz by oz by Oz 3: 2H 9 OH 2H, Q Ola Hs 9 OH Se OMe Or Or oz oz Or Or oz oz (9.4.19) Putting matrix [D] from Eq. (9.4.7) and matrix [B] from Eq. (9.4.19) into Eq. (9.1.22) computes the element stiffness matrix. [K*] = [B]" [DI[BV (9.4.20) where V is the volume of the tetrahedron element. 9.5 Axisymmetric Solid When the elasticity problem degenerates from three-dimension to axisymmetry, two shearing stress components vanish. These vanishing components due to symmetry are 7-9 and 7z9 in the r@z coordinate system where r is the radial direction, @ is the circumferential direction, and z is the axial direction. Hence, the remaining stress components are {o}={o, 09 0: Tz} (9.5.1) Similarly, the remaining strains are fe}={e € €: re} (9.5.2) The material property matrix [D] for the axisymmetric problem is l-v ov E vd [D] = —————. 9.5.3) 1p] (l+v)(1-2v)] v v (95:3) 0 0 The kinematic equation is & e 8 = = ig 9.5.4 & ge (9.5.4) ee eaeSection 9.5 Axisymmetric Solid 319 : L\- Figure 9.5.1 Free Body Diagram of Two-Dimensional Body “S (q1Y9) Pw ) : (py, where u and w are the radial and axial displacements, respectively. Example 9.5.1 Let us determine the circumferential strain ¢. If a hoop of radial r is uniformly displaced along the radial direction by displacement u. ‘Then, the deformed hoop has a uniform strain along the circumferential direction and the strain is computed as ex 2n(r+u)—2ar _u a : (9.5.5) In order to develop the element stiffness matrix, we use the linear triangular clement again. We substitute # and y in the shape functions, Eqs (5.2.7) through (5.2.9), with r and z for the axisymmetric problem. In addition, the axisymmetric element is a ring element as shown in Fig. 9.5.1. Substitution of the shape functions into the kinematic equation Eq. (9.5.4) gives Se Ey fy gk | =] > om bom bam | =D (9.5.6) on, 9H, oH, Ot, ons Oils — The element matrix can be expressed as [K*]= J ii J [B? [DI[B]d0 dr dz = 20 { [ “BIT [DI[Blar dz (9.5.7) Because of the term 4 in matrix [B], the matrix is not a constant matrix like the plane stress/strain case. As a result, the integration needs to be undertaken. One320 Elasticity Problem Chapter 9 simple approximation for the integration is to evaluate [B] at the centroid of the element. That is, we calculate #4) where # = S4t#t3 and z= 4422423. Then, the element stiffness matrix can be written as [K*] = 2aF AB] [DI[B) (9.5.8) in which [B] is the matrix [B] evaluated at the centroid of the element cross-section and A is its area. 9.6 Dynamic Analysis While previous sections consider static problems, this section considers dynamic problems. That is, we include the inertia force in equations of equilibrium. These equations are also called equations of motion. For the two-dimensional case, these equations are Ou O02 , Ot ane te +o Fz = Be + fe (9.6.1) fal a Oty [ Oey ot? Oz oy thy (9.6.2) where ¢ indicates time and p is the mass density. It can be easily extended to the three-dimensional case. Therefore, the finite element formulation for the dynamic problem contains one extra term compared to that for the static problem and the term is derived as follows using Galerkin’s method. [{mgh ae vf lfc a] (if eo (9.6.3) eal in which superimposed dot denotes temporal derivative. Using linear triangular elements, the accelerations can be interpolated as ty by #]_[m 0 m0 m ojJuel_, is ls Hy 0 Fa 0 Hs|)i, f= (96-4) tig 3 In Eq. (9.6.4) we assume that the shape functions are functions of spatial variables only and the nodal displacements are functions of time. Hence, the temporal derivative is performed for the nodal variable. Substituting Eq. (9.6.4) into Eq. (9.6.3) for each element results in fools a] fe} [ orrenenra (96.5)Section 9.7 MATLAB Application to 2-D Stress Analysis 321 As a result, the mass matrix is defined as we]= f owitiman (68) B This is the consistent mass matrix and it becomes for the linear triangular element 2) 0120.1 0 O20 01 _PA}1 02010 M= "51010201 (96.7) nO. 1022720 Ot 01 O22 This matrix is based on unit thickness of the element. Otherwise, it should be multiplied by the plate thickness. On the other hand, the lumped mass matrix for the linear triangular element is iu) = Ain (0.68) where [1] is the identity matrix of size 6. The consistent mass matrix for the bilinear element as shown in Fig. 5.3.1 is 4020 10 20 04020102 2 eee n»_pAlo 2040201 Ml=35]1 0204020 (9.6.9) 01020402 201020 40 720102 0 4 This matrix is based on unit thickness of the element. Otherwise, it should be multiplied by the plate thickness. The lumped mass matrix for this element is [M‘) = ean (9.6.10) where [J] is the identity matrix of size 8. 9.7 MATLAB Application to 2-D Stress Analysis Two-dimensional stress analyses are performed using both conventional finite elements and isoparametric elements in the following examples. 4 Example 9.7.1 A strip shown in Fig. 9.7.1 is subjected to an axial load. A MATLAB program is written to solve the problem using linear triangular elements. Eight elements are used as seen in the figure.322 Elasticity Problem Chapter 9 B=10°psi, v0.3 Figure 9.7.1 Plate Subjected to Axial Load %o % Example 9.7.1 % plane stress analysis of a solid using linear triangular elements % (see Fig. 9.7.1 for the finite element mesh) % % Variable descriptions % k = element matrix % { = element vector % kk = system matrix % ff = system vector % disp = system nodal displacement vector % eldisp = element nodal displacement vector % stress = matrix containing stresses % strain = matrix containing strains % gcoord = coordinate values of each node % nodes = nodal connectivity of each element % index = a vector containing system dofs associated with each element % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in bedof % % G——__ ——. % input data for control parameters % number of elements % number of nodes per element % number of dofs per node % total number of nodes in system % total system dofs % degrees of freedom per element % elastic modulus poisson=0.3; % Poisson’s ratioSection 9.7 MATLAB Application to 2-D Stress Analysis % 6——$$ $$$ __________— % input data for nodal coordinate values % gcoord(ij) where i-> node no. and j-> x ory (hl gcoord=[0.0 0.0; 0.0 1.0; 1.0 0.0; 1.0 1. 5 2.0 1.0; 3.0 0.0; 3.0 1.0; 4.0 0.0; 4.0 1.0}; % % % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes %— ee nodes=[1 3 4; 14 2; 356; 36 4; 57 8; 5 8 6;7 9 10; 7 10 8); % Cus % input data for boundary conditions % first three dofs are constrained % whose described values are 0 % initialization of matrices and vectors oo f=zeros(sdof,1); % system force vector kk=zeros(sdof,sdof); % system matrix i % system displacement vector % element displacement vector % matrix containing stress components % matrix containing strain components % index vector % kinematic matrix matmtx=zeros(3,3); ‘% constitutive matrix %— % force vector % force applied at node 9 in x-axis % force applied at node 10 in x-axis matmtx=fematiso(1,emodule, poisson); % constitutive matrix % for iel=1:nel % loop for the total number of elements % nd(1)=nodes(iel,1); % Ast connected node for (iel)-th element 323324 Elasticity Problem Chapter 9 nd(2)=nodes(iel,2); % 2nd connected node for (iel)-th element nd(3)=nodes(iel,3); % ard connected node for (iel)-th element % x1=geoord(nd(1),1); yl=gcoord(nd(1),2); -% coord values of Ist node x2=gcoord(nd(2),1); y2=gcoord(nd(2),2); -% coord values of 2nd node x3=gcoord(nd(3),1); y3=gcoord(nd(3),2); % coord values of 3rd node ” index=feeldof(nd,nnel,ndof); % extract system dofs for the element % ~%— % find the derivatives of shape functions 6. area=0.5*(x1*y2+x2*y3+x3*y 1-x1*y3-x2*y1-x3*y2); % area of triangule area*2; ‘1 /area2)*[(y2-y3) (y3-y1) (y1-y2)]; % derivatives w.r.t. x dhdy=(1/area2)*[(x3-x2) (x1-x3) (x2-x1)}; % derivatives w.t.t. y % kinmtx2=fekine2d(nnel,dhdx,dhdy); % kinematic matrix % k=kinmtx2”*matmtx*kinmtx2*area; % element stiffnes matrix % kk=feasmbl1(kk,k,index); % assemble element matrices % end % end of loop for the total number of elements % oa % apply boundary conditions kk, ff]=feaplyc2({ck,ff,bedof, beval); ” %—_________— % solve the matrix equation %——__—___—__—. disp=kk\ ff % 6 % element stress computation (post computation) (eee ees for ielp=1:nel % loop for the total number of elements % nd(1)=nodes(ielp,1); % Ist connected node for (iel)-th element odes(ielp,2); % tnd connected node for (iel)-th element odes(ielp,3); % 8rd connected node for (iel)-th element oord(nd(1),1); yl=gcoord(nd(1),2); -% coord values of Ist node gcoord(nd(2),1); y2=gcoord(nd(2),2); % coord values of 2nd node x3=gcoord(nd(3),1); y3=gcoord(nd(3),2); % coord values of 3rd node %Section 9.7 MATLAB Application to 2-D Stress Analysis 325 index=feeldof(nd,nnel,ndof); % extract system dofs for the element. % ee % extract element displacement vector Oe for eldisp(i end % area=0.5*(x1*y2+x2*y3-+x3*y1-x1*y3-x2*yl-x3*y2); % area of triangle area2=area*2; of isp(index(i)); dhdx=(1/area2)*[(y2-y3) (y3-y1) (yl-y2)]; % derivatives w.r.t. x dhdy=(1/area2)*[(x3-x2) (x1-x3) (x2-x1)]; % derivatives wart. y % kinmtx2=fekine2d (nnel,dhdx,dhdy); % kinematic matrix % estrain=kinmtx2*eldisp; % compute strains estress=matmtx*estrain; % compute stresses % for strain(ielp,i % store for each element stress(ielp,i)=estress(i); % store for each element end % end % %— % print fem solutions % print nodal displacements stresses~[i stress(i,:)] % print stresses end % % function [kinmtx2]=fekine2d(nnel,dhdx,dhdy) % % Purpose: % determine the kinematic equation between strains and displacements % for two-dimensional solids % % Synopsis % (kinmtx2]=fekine2d(nnel,dhdx,dhdy)326 Elasticity Problem Chapter 9 % % Variable Description: % nnel - number of nodes per element % dhdx - derivatives of shape functions with respect to x % dhdy - derivatives of shape functions with respect to y kinmtx2(3,i2)=dhdx(i); end % — function [matmtrx]=fematiso(iopt,elastic,poisson) %- % Purpose: % determine the constitutive equation for isotropic material % % Synopsis: % {matmtrx]=fematiso(iopt,elastic,poisson) % % Variable Description: ‘% elastic - elastic modulus % poisson - Poisson’s ratio - plane stress analysis ~ plane strain analysis - axisymmetric analysis % plane stress elastic/(1-poisson*poisson)* [1 poisson 0: poisson 1 0; ... 0 0 (1-poisson) /2]; % elseif iopt==2 % plane strain matmtrx= elastic/((1+poisson)*(1-2*poisson))* ... [(1-poisson) poisson 0; poisson (1-poisson) 0; 0 0 (1-2*poisson)/2];Section 9.7 lastic/((1+poisson)*(1-2*poisson))* ... [(1-poisson) poisson poisson 0; poisson (1-poisson) poisson 0; poisson poisson (1-poisson) 0; 0 00 (1-2*poisson)/2]; % else matmtrx= elastic/((1+poisson)*(1-2*poisson))* ... [(-poisson) poisson poisson 0 0 0; poisson (1-poisson) poisson 0 0 0; poisson poisson (1-poisson) 0 0 0; 0 0.0 (1-2*poisson)/2 0 0; 0.0 00 (1-2*poisson)/2 0; 0.000 0 (1-2*poisson)/2]; % end % %- MATLAB Application to 2-D Stress Analysis % axisymmetry % three-dimension ‘The nodal displacements are listed below and they agree with the analytical solutions. On the other hand, the state of stress of each element is cz = 1000 and oy = Tey = 0 as expected. displace = dof, displ. 1.0000 0.0000 % x-displacement of node 1 2.0000 0,000 % y-displacement of node 1 3.0000 0.0000 % x-displacement of node 2 4.0000 —--0.0030 % y-displacement of node 2 5.0000 0.0100 % x-displacement of node 3 6.0000 0.0000 % y-displacement of node 3 7.0000 0.0100 % x-displacement of node 4 8.0000 -0.0030 % y-displacement of node 4 9.0000 0.0200 % x-displacement of node 5 10.000 —-0,0000 % y-displacement of node 5 11.000 0.0200 % x-displacement of node 6 12.000 -0.0030 % y-displacement of node 6 13.000 0.0300 % x-displacement of node 7 14.000 0.0000 % y-displacement of node 7 15.000 0.0300 % x-displacement of node 8 16.000 0.0030 % y-displacement of node & 17.000 0.0400 % x-displacement of node 9 18.000 0,000 % y-displacement of node 9 19.000 0.0400 % x-displacement of node 10 20,000 -0.0030 % y-displacement of node 10 327328 Elasticity Problem Chapter 9 Ftp, »=03 Figure 9.7.2 Cantilever Beam Subjected to a Tip Load Example 9.7.2 We want to analyze a short cantilever beam using two- dimensional isoparametric elements assuming plane stress condition. To this end, the beam is modeled using ten four-node quadrilateral elements as seen in Fig. 9.7.2. % % Example 9.7.2 % plane stress analysis of a cantilever beam using isoparametric % four-node elements % (see Fig. 9.7.2 for the finite element mesh) % % Variable descriptions % k = element matrix % { = element vector % kk = system matrix system vector system nodal displacement vector % eldisp = element nodal displacement vector % strain = matrix containing strains % gcoord = coordinate values of each node % nodes = nodal connectivity of each element % index = a vector containing system dofs associated with each element matrix containing sampling points % weight2 = matrix containing weighting coefficients % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in bedof % % %- % input data for control parametersSection 9.7 MATLAB Application to 2-D Stress Analysis % number of elements % number of nodes per element % number of dofs per node nnode=18; % total number of nodes in system sdof=nnode*ndof; % total system dofs edof=nnel*ndof; % degrees of freedom per element emodule=le6; % clastic modulus % Poisson’s ratio % 2x2 Gauss-Legendre quadrature nglxy=nglx*ngly; % number of sampling points per element % fo % input data for nodal coordinate values % gcoord(i,j) where i-> node no. and j-> x or y el gcoord=[0.0 0.0; 0.0 1.0; 0.5 0.0; 0.5 1.0; 1.0 0.0; 1.0 1.0; 1.5 0.0; 1.5 1.0; 2.0 0.0; 2.0 1.0; 2.5 0.0; 2.5 1.0; 3.0 0.0; 3.0 1.05 3.5 0.0; 3.5 1.0; 4.0 0.0; 4.0 1.0); % eee % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes %— a — nodes=[1 3 42; 35 6 4;578 6; 79 108; 9 11 12 10; 11 13 14 12; 13 15 16 14; 15 17 18 16]; % ——— % input data for boundary conditions % first four dofs are constrained % whose described values are 0 % system force vector % system matrix disp=zeros(sdof,1); % system displacement vector eldisp=zeros(edof,1); % element displacement vector stress=zeros(nglxy,3); % matrix containing stress components strain=zeros(nglxy,3); % matrix containing strain components index=zeros(edof,1); % index vector kinmtx=zeros(3,edof); % kinematic matrix matmtx=zeros(3,3); % constitutive matrix % %——__—_—— 329330 Elasticity Problem Chapter 9 % force vector oo f1(34)=500; % force applied at node 17 in y-axis fi(36)=500; % force applied at node 18 in y-axis % %— a % compute element matrices and vectors, and assemble %— {point2,weight2]=feglqd2(nglx,ngly); % sampling points & weights matmtx=fematiso(1,emodule, poisson); % constitutive matrix % for iel=1:nel % loop for the total number of elements % extract nodes for (iel)-th element % extract x value of the node % extract y value of the node k=zeros(edof,edof); % initialization of element matrix % oo % numerical integration %——_—. for intx=I:nglx x=point2(intx,1); % sampling point in x-axis wtx=weight2(intx,1); % weight in x-axis for inty=I:ngly y=point2(inty,2); % sampling point in y-axis wty=weight2(inty,2) ; % weight in y-axis % [shape,dhdr,dhds]=feisoq4(x,y); % compute shape functions and % derivatives at sampling point % jacob2=fejacob2(nnel,dhdr,dhds,xcoord,ycoord); _% compute Jacobian % detjacob=det(jacob2); % determinant of Jacobian inyjacob=inv(jacob2); % inverse of Jacobian matrix % [dhdx,dhdy]=federiv2(nnel,dhdr,dhds,invjacob); % derivatives w.r.t. % physical coordinate % kinmtx2=fekine2d(nnel,dhdx,dhdy); % compute kinematic matrix % Cea % compute element matrix goatee k=k-++kinmtx2”*matmtx*kinmtx2*wtx*wty*detjacob; % element matrixSection 9.7 MATLAB Application to 2-D Stress Analysis % end end % end of numerical integration loop % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % kk=feasmbl1 (kk,k,index); % assemble element matrices % end % end of loop for the total number of elements % ————— % apply boundary conditions i —_—_--—_—_— [kk, ff] =feaplyc2(kk,ff, bedof, beval); q ee % solve the matrix equation %— disp=kk\ ff % num=1:1:sdof; displace=[num’ disp] % print nodal displacements —%— % element stress computation ——————_—____—_—_——— for ielp=1:nel % loop for the total number of elements nd(i)=nodes(ielp,i); % extract nodes for (iel)-th element xcoord(i)=gcoord(nd(i),1); % extract x value of the node yeoord(i)=gcoord(nd(i), % extract y value of the node end % oo % namerical integration for intx=I:nglx x=point2(intx,1); % sampling point in x-axis wtx=weight2(intx,1); % weight in x-axis for inty=I:ngly y=point2(inty,2); % sampling point in y-axis wty=weight2(inty,2) ; % weight in y-axis intp=intp+1; % {shape,dhdr,dhds]=feisoq4(x,y); % compute shape functions and % derivatives at sampling point 331332 Elasticity Problem Chapter 9 % jacob2=fejacob2(nnel,dhdr,dhds,xcoord,ycoord); _% compute Jacobian % detjacob=det(jacob2); % determinant of Jacobian inyjacob=inv(jacob2); % inverse of Jacobian matrix % [dhdx,dhdy]=federiv2(nnel,dhdr,dhds,invjacob); % derivatives w.r.t. % physical coordinate % kinmtx2=fekine2d(nnel,dhdx,dhdy); % kinematic matrix % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % %— % extract element displacement vector = for i=l:edof eldisp(i)=disp(index(i)); end % kinmtx2=fekine2d(nnel,dhdx,dhdy); % compute kinematic matrix % compute strains estress: i % compute stresses for i: strain(intp,i)=estrain(i); % store for each sampling point stress(intp,i)=estress(i); % store for each sampling point end % location=[ielp,intx,inty] % print location for stress stress(intp,:) % print stress values % end end % end of integration loop % end % end of loop for total number of elements % —%— ‘As expected, the displacements in the z-axis are positive at the bottom side of the strip and negative at the top side because of bending. The tip displacement in the y-axis is 0.2238 in. while the beam bending theory gives 0.256. As a result, the mesh needs to be refined to improve the accuracy. On the other hand, the bending stress oz is 11950 psi at the integration point nearest to the fixed edge. The point is located 0.1057 in, away from the fixed edge in the z-axis and 0.2887 in. above from the midplane in the y-axis. The beam theory gives bending stress of 13820.Section 9.7 MATLAB Application to 2-D Stress Analysis displace = dof. displ. 1.0000 0.0000 2.0000 0.0000 3.0000 0.0000 4,0000 0.0000 5.0000 0.0094 6.0000 0.0060 7.0000 -0.0094 8.0000 0.0060 9.0000 0.0176 10.000 0.0208 11.000 -0.0176 12.000 0.0208 13.000 0.0245 14.000 0.0431 15,000 -0.0245 16.000 0.0431 17.000 0.0301 18.000 0.0717 19.000 -0.0301 20.000 © 0.0717 21,000 0.0345 22.000 0.1534 23,000 -0.0345 24.000 (0.1053 25,000 0.0377 26.000 0.1427 27.000 -0.0377 28.000 0.1427 29.000 0.0395 30.000 0.1826 31.000 -0.0395 32.000 0.1826 33.000 0.0402 34.000 0.2238 35.000 -0.0402 36.000 0.2238 % x-displacement of node 1 % y-displacement of node 1 % x-displacement of node 2 % y-displacement of node 2 % x-displacement of node 3 % y-displacement of node 3 % x-displacement of node 4 % y-displacement of node 4 % x-displacement of node 5 % y-displacement of node 5 % x-displacement of node 6 % y-displacement of node 6 % x-displacement of node 7 % y-displacement of node 7 % x-displacement of node 8 % y-displacement of node 8 % x-displacement of node 9 % y-displacement of node 9 % x-displacement of node 10 % y-displacement of node 10 % x-displacement of node 11 % y-displacement of node 11 % x-displacement of node 12 % y-displacement of node 12 % x-displacement of node 13 % y-displacement of node 13 % x-displacement of node 14 % y-displacement of node 14 % x-displacement of node 15 % y-displacement of node 15 % x-displacement of node 16 % y-displacement of node 16 % x-displacement of node 17 % y-displacement of node 17 % x-displacement of node 18 % y-displacement of node 18 333334 Elasticity Problem Chapter 9 fh B20" psi v =0.25 Figure 9.8.1 Axisymmetric Solid With Triangular Elements 9.8 MATLAB Application to Axisymmetric Analysis The same axisymmetric solid is analyzed using both conventional triangular elements and isoparametric quadrilateral elements. & Example 9.8.1 A thick walled cylinder is subjected to a uniform internal pressure of 2000 psi. The cylinder has inside radius of 10 in. and outside radius of 15 in. It is made of steel whose elastic modulus is 28x10® psi and Poisson’s ratio is 0.3. Figure 9.8.1 shows the finite element mesh using 10 triangular elements. For the present analysis, the cylinder is assumed to be constrained along the axial direction. The resultant force applied on the inside surface is 2000 x 2x x 10 = 20,0007 assuming unit length along the axial direction. As a result, nodes 1 and 2 at the inside boundary has a concentrated nodal force of 10, 000m along the radial direction, respectively. The value is provided to the force vector in the finite element analysis program. % % Example 9.8.1 % axisymmetric analysis of a solid subjected to an internal % pressure using linear triangular elements % (see Fig. 9.8.1 for the finite element mesh) % % Variable descriptions % k = element matrix % { = element vector % kk = system matrix % ff = system vector % disp = system nodal displacement vector % eldisp = element nodal displacement vectorSection 9.8 MATLAB Application to Axisymmetric Analysis 335 % stress = matrix containing stresses % strain = matrix containing strains % gcoord = coordinate values of each node % nodes = nodal connectivity of each element % index = a vector containing system dofs associated with each element % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in bcdof % % %- % input data for control parameters % % number of elements % number of nodes per element % number of dofs per node % total number of nodes in system % total system dofs % degrees of freedom per element emodule=28e6; % elastic modulus poisson=0.25; % Poisson’s ratio % Go % input data for nodal coordinate values % gcoord(ij) where i-> node no. and j-> x or y %———__—_. EES gcoord=[10. 0.; 10. 15 11. 0. 11. 15 12. 05 12. 15 oo ete te Oe ee ee % % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes —-—————— cq nodes=[1 34; 142; 35 6; 3645578; 5 8 6;7 9 10; 7 10 8; 9 11 12; 9 12 10); % ge % input data for boundary conditions bedof=[2 4 6 8 10 12 14 16 18 20 22 24); % axial motion constrained beval=[0 00000000000); % constrained values are 0 % %— ———. % initialization of matrices and vectors %——___—__—_____——. ff=zeros(sdof,1); % system force vector kk=zeros(sdof,sdof); % system matrix336 Elasticity Problem Chapter 9 disp=zeros(sdof,1); % system displacement vector eldisp=zeros(edof,1); % element displacement vector stress=zeros(nel,4); % matrix containing stress components strain=zeros(nel,4); % matrix containing strain components index=zeros(edof,1); % index vector kinmtax=zeros(4,edof); % kinematic matrix matmtx=zeros(4,4); % constitutive matrix % % % force vector % .0*atan(1); % pi=3.141592 % ff(1)=2e3*pi*2*10; % force applied at node 1 in x-axis ff(3)=2e3*pit2*10; % force applied at node 2 in x-axis % — matmtx=fematiso(3,emodule,poisson); % constitutive matrix % for iel=1:nel % \oop for the total number of elements % nd(1)=nodes(iel,1); % Ist node for (iel)-th element nd(2)=nodes(iel,2); % 2nd node for (iel)-th element nd(3)=nodes(iel,3); % 3rd node for (iel)-th element % x1=geoord(nd(1),1); yl=geoord(nd(1),2); -% coord values of Ist node x2=gcoord(nd(2),1); y! ‘oord(nd(2),2); % coord values of 2nd node x3=gcoord(nd(3),1); y8=gcoord(nd(3),2); % coord values of 3rd node % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % % % find the derivatives of shape functions %— area=0.5*(x1*y2-+x2*y3-+x3*y1-x1*y3-x2*y1-x3*y2); % area of triangle area2=area*2; xcenter=(x1-+x2+x3)/3; % x-centroid of triangle yeenter=(yl+y2+y3)/3; % y-centroid of triangle % shape(1)=((x2*y3-x3*y2)+(y2-y3)*xcenter+(x3-x2)*ycenter) /area2; shape(2)=((x3*y1-x1*y3)+(y3-y1)*xcenter+(x1-x3)*ycenter) /area2; shape(3)=((x1*y2-x2*y1)+(yl-y2)*xcenter-+(x2-x1)*ycenter) /area2; % dhdx=(1/area2)*({(y2-y3) (y3-y1) (y1-y2)}; % derivatives w.r.t. x dhdy=(1/area2)*[(x3-x2) (x1-x3) (x2-x1)]; % derivatives w.r.t. y %Section 9.8 MATLAB Application to Axisymmetric Analysis 337 kinmtax=fekineax(nnel,dhdx,dhdy,shape,xcenter); _% kinematic matrix % k=2*pi*xcenter*area*kinmtax’*matmtx*kinmtax; _-% element matrix % kk=feasmbl1 (kk,k,index); % assemble element matrices % end % end of loop for total number of elements % ——— % apply boundary conditions ge oa {kk, ff] =feaplyc2(kk, ff, bedof, beval); % —————— % solve the matrix equation % disp=kk\ fi; % %——— —— % clement stress computation (post-computation) (ee for iel) el % loop for the total number of elements % nd(1)=nodes(ielp,1); % 1st node for (iel)-th element nd(2)=nodes(ielp,2); % 2nd node for (iel)-th element nd(3)=nodes(ielp,3); % 3rd node for (iel)-th element % coord(nd(1),2); % coord values of 1st node coord(nd(2),2); % coord values of 2nd node x3=geoord(nd(3),1); y3=gcoord(nd(3),2); _% coord values of 3rd node % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % ee % extract element displacement vector % Se for i=1:edof eldisp(i)=disp(index(i)); end % area=0.5*(x1*y2+x2*y3-+4x3*yl-x1l*y3-x2*y1-x3*y2); % area of triangle area2=area*2; x1+x24x3)/3; % x-centroid of triangle yeenter=(y1+y2-+y3)/3; % y-centroid of triangle % shape(1)=((x2*y3-x3*y2)+(y2-y3)*xcenter-+(x3-x2)*ycenter)/area2; shape(2)=((x3*yl-x1*y3)+(y3-y1)*xcenter+(x1-x3)*ycenter) /area2; shape(3)=((x1*y2-x2*y1)+(y1-y2)*xcenter-+(x2-x1)*ycenter)/area2;Elasticity Problem Chapter 9 1/area2)*[(y2-y3) (y3-y1) (y1-y2)}; % derivatives w.r.t. x 1 /area2)*((x3-x2) (x1-x3) (x2-x1)}; % derivatives w.r.t. y kinmtax=fekineax(nnel,dhdx,dhdy,shape,xcenter); % kinematic matrix % compute strains % compute stresses strain (ielp,i)=estrain(i); % store for each element stress(ielp,i)=estress(i); % store for each element end % end % %- % print fem solutions —b——_——— :sdof; displace=[num’ disp] % print nodal displacements num=: % print stresses function [kinmtxax]=fekineax(nnel,dhdx,dhdy,shape,tadist) %— % Purpose: % determine kinematic equations between strains and displacements % for axisymmetric solids % % Synopsis: % (kinmtxax]=fekineax(nnel,dhdx,dhdy,shape,radist) % % Variable Description: % mel - number of nodes per element % shape - shape functions % dhdx - derivatives of shape functions with respect to x % dhdy - derivatives of shape functions with respect to y % radist - radial distance of integration point or central point % for hoop strain component % %Section 9.8 MATLAB Application to Axisymmetric Analysis for i=1:nnel i1=(i-1)*241; i2=i141; kinmtxax(1,i1)=dhdx(i); kinmtxax(2, kinmtxax(3,i2 kinmtxax(4,i1)=dhdy(i); kinmtxax(4,i2)=dhdx(i); end % %— ‘The results are displace dof. displ. 1.0000 9.0039 % radial displacement of node 1 2.0000 90,0000 3.0000 0.0039 % radial displacement of node 2 4.0000 0.0000 5.0000 0.0037 % radial displacement of node 3 6.0000 0.0000 7.0000 0.0037 % radial displacement of node 4 8.0000 0.0000 9.0000 0,035 % radial displacement of node 5 10.000 0.0000 11.000 0.0035 % radial displacement of node 6 12.000 9.0000 13.000 0.0034 % radial displacement of node 7 14.000 0.0000 15.000 0.0034 % radial displacement of node 8 16.000 0.0000 17.000 0.0033 % radial displacement of node 9 18.000 0.0000 19.000 0.0033 % radial displacement of node 10 20.000 9.0000 21.000 0.0032 % radial displacement of node 11 22.000 0.0000 23.000 0.0032 % radial displacement of node 12 24.000 0.0000 stresses = elem radial © hoop -—axial 1.00 -3277, 9486. 1552. 2.00 -3408. 100001649. 300-2281. 8434. «1538. 4.00 -2221. 8877. 1664. 500-1457. 7637. «1545. 6.00 -1368. 7995. 1657. 339340 Elasticity Problem Chapter 9 B=2.8X10 psi v=0.25 Figure 9.8.1 Axisymmetric Solid With Rectangular Elements 7.00 -783.0 7017. «1558. 8.00 -724.0 7295. 1643. 9.00 213.0 6529. 1579. 10.0 -242.0 6727. 1621. & Example 9.8.2 Example 9.8.1. is solved using isoparametric elements. ‘The same number of nodes is used but the number of element is 5 as seen in Fig. 9.8.2 %— % Example 9.8.2 % axisymmetric analysis of a thick walled cylinder % subjected to internal pressure using isoparametric % four-node elements % (see Fig. 9.8.2 for the finite element mesh) % % Variable descriptions % k = element matrix element vector system matrix system vector % disp = system nodal displacement vector % eldisp = element nodal displacement vector % stress = matrix containing stresses % strain = matrix containing strains % gcoord = coordinate values of each nodeSection 9.8 MATLAB Application to Axisymmetric Analysis 341 % nodes = nodal connectivity of each element % index = a vector containing system dofs associated with each element % point2 = matrix containing sampling points matrix containing weighting coefficients a vector containing dofs associated with boundary conditions a vector containing boundary condition values associated with the dofs in bedof % input data for control parameters % number of elements % number of nodes per element % number of dofs per node % total number of nodes in system % total system dofs nel*ndof; % degrees of freedom per element emodule=28.0e6; % elastic modulus poisson=0.2: % Poisson’s ratio nglx=2; ngly=2; % 2x2 Gauss-Legendre quadrature nglxy=nglx*ngly; % number of sampling points per element ————— % input data for nodal coordinate values % gcoord(i,i) where i-> node no. and j-> x or y —————————————— geoord=[10. 0.5 10. 15 11. 0.5 11. 13, 05 13. 15 14. 05 14. 15 15. 0,5 15. 1); % ee % input data for nodal connectivity for each element % nodes(i,j) where i-> element no. and j-> connected nodes % nodes=[1 3 4 2; 35 6 4; 5 78 6; 79 10 8; 9 11 12 10} % ee % input data for boundary conditions eS ee bedof=[2 4 6 8 10 12 14 16 18 20 22 24); % axial motion constrained beval=[0 00000000000); % constrained values are 0 % initialization of matrices and vectors ros(sdof,1); % system force vector eros(sdof,sdof); % system matrix -zeros(sdof,1); % system displacement vector342 Elasticity Problem Chapter 9 eldisp=zeros(edof,1); % element displacement vector stress=zeros(nglxy,4); % matrix containing stress components strain=zeros(nglxy,4); % matrix containing strain components index=zeros(edof,1); % index vector kinmtx=zeros(4,edof); % kinematic matrix matmtx=zeros(4,4); % constitutive matrix % [ace % force vector Os pi=4.0*atan(1.0); % pi=3.141592 % $€(1)=2e3*2*pi*10; % force applied at node 1 in x-axis ff(3)=2e3*2* pi*10; % force applied at node 2 in x-axis % %— % compute element matrices and vectors, and assemble %— [point2, weight2]=feglqd2(nglx,ngly); % sampling points & weights matmtx=fematiso(3,emodule, poisson); % constitutive matrix % for iel=1:nel % \oop for the total number of elements % extract node for (iel)-th element % extract x value of the node yeoord(i % extract y value of the node end % k=zeros(edof,edof); % initialization of element matrix % %— % numerical integration %— _——_——_— for intx=1:nglx x=point2(intx,1); % sampling point in x-axis wtx=weight2(intx,1); % weight in x-axis for inty=I:ngly y=point2(inty,2); % sampling point in y-axis wty=weight2(inty,2) ; % weight in y-axis % [shape,dhdr,dhds]=feisoad(x,y); % compute shape functions and % derivatives at sampling point % jacob2=fejacob2(nnel,dhdr,dhds,xcoord,ycoord); _% compute Jacobian % determinant of Jacobian % inverse of Jacobian matrixSection 9.8 MATLAB Application to Axisymmetric Analysis % [dhdx,dhdy]=federiv2(nnel,dhdrdhds,invjacob); _-% derivatives w.r.t. % physical coordinate ‘% x-coordinate value xcenter=xcenter+shape(i)*xcoord(i); % of the integration point end % kinmtx=fekineax(nnel,dhdx,dhdy,shape,xcenter); _% kinematic matrix % ra % compute element matrix eee k=k+2*pi*xcenter*kinmtx’*matmtx*kinmtx*wtx* wty*detjacob; % element matrix % end end % end of numerical integration loop % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % kk=feasmbl1(kk,k,index); % assemble element matrices % end % end of loop for total number of elements % —— % apply boundary conditions $—__________—- [kk ff] =feaplyc2(kk, ff,bedof, beval); % ——— % solve the matrix equation eee disp=kk\ ff; % num=: dof, displace={num’ disp] % print nodal displacements % % % element stress computation % loop for the total number of elements % extract node for (iel)-th element % extract x value of the node % extract y value of the node 343344 Elasticity Problem end % % % numerical integration x=point2(intx,1); wtx=weight2(intx,1); for inty=I:ngly y=point2(inty,2); wty=weight2(inty,2) ; intp=intp+1; % [shape,dhdr,dhds]=feisoa4(x,y); % jacob2=fejacob2(nnel,dhdr,dhds,xcoord,ycoord); % detjacob=det(jacob2); invjacob=inv(jacob2); % [dhdx,dhdy]=federiv2(nnel,dhdr,dhds,invjacob); for i=I:nnel xcenter=xcenter+shape(i)*xcoord(i); end % kinmtx=fekineax(nnel,dhdx,dhdy,shape,xcenter); % index=feeldof(nd,nnel,ndof); % —————————————_———— ‘% extract element displacement vector (ee for i=1:edof eldisp(i)=disp(index(i)); inmtx*eldisp; atmtx*estrain; strain(intp,i)=estrain(i); stress(intp,i)=estress(i); end Chapter 9 % sampling point in x-axis % weight in x-axis % sampling point in y-axis % weight in y-axis % compute shape functions and ‘% derivatives at sampling point % compute Jacobian % determinant of Jacobian % inverse of Jacobian matrix ‘% derivatives w.r.t. % physical coordinate % x-coordinate value % of the integration point % kinematic matrix % extract system dofs for the element % compute strains % compute stresses % store for each element % store for each elementSection 9.9 MATLAB Application to 3-D Stress Analysis % end end % end of integration loop % for j=1:nglxy stresses: end % end ‘% end of loop for total number of elements % % ielp stress(j,:)] % print stresses ‘The nodal displacements are the same as those obtained from Example 9.8.1. ‘The stresses are also very similar but a little different. Stresses for the first two elements are listed below. They are printed for each integration point. Because 2x2 quadrature was used for numerical integration, there are four integration points for each element. stresses for element 1 = Ist integration point radial stress=-3146. hoop stress=10289 axial stress=1786 2nd integration point radial stress=-3146. hoop stress=10289 axial stress=1786. 3rd integration point radial stress=-3505. hoop stress=9211. axial stress=1426. 4th integration point radial stress=-3506. hoop stress=9211. axial stress=1426. stresses for element 2 = Ast integration point radial stress=-2104. hoop stress=9066. axial stress=1741. 2nd integration point radial stress=-2104. hoop stress=9066. axial stress=1741. 3rd integration point radial stress=-2377. hoop stress=8245. axial stress=1467. Ath integration point radial stress=-2377. hoop stress=8245. axial stress=1467. 9.9 MATLAB Application to 3-D Stress Analysis @ Example 9.9.1 A unit cube is subjected to a uniform load as shown in Fig. 9.9.1, We use one eight-node isoparametric solid element for the problem. % % Example 9.9.1 345346 Elasticity Problem Chapter 9 a (O11). Gant) (0,01, (4,1,0) (0,0,0) (1,0,0) Figure 9.9.1 A Cubic Solid % three-dimensional analysis of a cube using isoparametric % eight-node elements % (see Fig. 9.9.1 for the finite element mesh) % % Variable descriptions % k = element matrix % £ = element vector % kk = system matrix % ff = system vector % disp = system nodal displacement vector % eldisp = element nodal displacement vector % stress = matrix containing stresses % strain = matrix containing strains % gcoord = coordinate values of each node % nodes = nodal connectivity of each element % index = a vector containing system dofs associated with each element % point3 = matrix containing sampling points % weight3 = matrix containing weighting coefficients % bcdof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in bedof % % —EE % input data for control parameters % nel: % number of elements anel % number of nodes per element ndof: % number of dofs per node nnode=8; % total number of nodes in system sdof=nnode*ndof; % total system dotsSection 9.9 MATLAB Application to 3-D Stress Analysis edof=nnel*ndof; % degrees of freedom per element % elastic modulus % Poisson’s ratio 5 nglz=: % 2x2x2 Gauss-Legendre quadrature nglxyz=nglx*ngly*ngl; % number of sampling points per element % 7 % input data for nodal coordinate values % gcoord(i,j) where i-> node no. and j-> x or y (ee gcoord=[0.0 0.0 0.0; 1.0 0.0 0.0; 1.0 1.0 0.0; 0.0 1.0 0.0; 0.0 0.0 1.0; 1.0 0.0 1.0; 1.0 1.0 1.0; 0.0 1.0 1.0}; % 6 % input data for nodal connectivity for each element % nodes(i) where i-> element no. and j-> connected nodes nodes=[1 2 3 45 6 7 8); % A —- % input data for boundary conditions Ce bedof=[1 23 56 9 12}; 0000000); °—————_————_ % initialization of matrices and vectors %— a fi=zer0s(sdof,1); ros(sdof,sdof); disp=zeros(sdof,1); % constrained dofs ‘% constrained values % system force vector % system matrix % system displacement vector eldisp=zeros(edof,1); % element displacement vector % matrix containing stress components % matrix contai g strain components % index vector % kinematic matrix % constitutive matrix % force applied at node § in z-axis % force applied at node 6 in z-axis % force applied at node 7 in z-axis % force applied at node 8 in z-axis % compute element matrices and vectors, and assemble 347348 Elasticity Problem Chapter 9 ———— $$$ [point3,weight3]=feglqd3(nglx,nglynglz); _% sampling points & weights matmtx=fematiso(4,emodule,poisson); % compute constitutive matrix % for iel=1:nel % loop for the total number of elements % extract node for (iel)-th element % extract x value of the node ‘% extract y value of the node zcoord(i)=gcoord(nd(i),3); % extract z value of the node end % k=zeros(edof,edof); % initialization of element matrix % %— % numerical integration %— for intx=I:nglx x=point3(intx,1); % sampling point in x-axis wtx=weight3(intx,1); % weight in x-axis for inty=l:ngly y=point3(inty,2); % sampling point in y-axis wty=weight3(inty,2) ; % weight in y-axis for intz=1:nglz 2=point3(intz,3); % sampling point in z-axis wtz=weight3(intz,3) ; % weight in z-axis % [shape,dhdr,dhds dhdt]=feisos8(x,y,z); % compute shape functions % and derivatives at sampling point % jacob3=fejacob3(nnel,dhdr,dhds,dhdt,xcoord,ycoord,zcoord); % compute Jacobian % determinant of Jacobian % inverse of Jacobian matrix [dhdx,dhdy,dhdz]=federiv3(nnel,dhdr,dhds,dhdt,invjacob); % derivatives w.r.t. physical coordinate % kinmtx=fekine3d(nnel,dhdx,dhdy,dhdz); % kinematic matrix % %— ———__—— % compute element matrix je k=k-+kinmtx’*matmtx*kinmtx*wtx*wty*wtz*detjacob; % element % endSection 9.9 MATLAB Application to 3-D Stress Analysis end end % end of numerical integration loop % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % kk=feasmbl1(kk,k,index); % assemble element matrices % end % end of loop for total number of elements % a % apply boundary conditions &¢——$________ kk, ff] =feaplyc2(kk,ff, bedof, beval); % % % solve the matrix equation anima disp=kk\ ft displace=[num’ disp] % print nodal displacements % % % element stress computation (ee for ielp=1:nel % loop for the total number of elements odes(ielp,i); % extract node for (iel)-th element xcoord(i)=gcoord(nd(i),1); % extract x value of the node % extract y value of the node % extract z value of the node Linglx x=point3(intx,1); % sampling point in x-axis wtx=weight3(intx,1); % weight in x-axis for inty=1:ngly y=point3(inty,2); % sampling point in y-axis wty=weight3(inty,2) ; % weight in y-axis for intz=1:nglz z=point3(intz,3); % sampling point in z-axis wtz=weight3{intz,3) ; % weight in z-axis intp=intp+1; 349350 Elasticity Problem Chapter 9 % [shape,dhdr,dhds,dhdt]=feisos8(x,y,2); % compute shape functions % and derivatives at sampling point % jacob3=fejacob3(nnel,dhdr,dhds,dhdt,xcoord, ycoord,zcoord); % compute Jacobian % detjacob=det(jacob3); % determinant of Jacobian invjacob=inv(jacob3); % inverse of Jacobian matrix % [ahdx,dhdy,dhdz}=federiv3 (nnel,dhdr,dhds,dhdt,invjacob); % derivatives w.r.t. physical coordinate % kinmtx=fekine3d(nnel,dhdx,dhdy,dhdz); -% compute kinematic matrix % index=feeldof(nd,nnel,ndof); % extract system dofs for the element % %—__—_____—— % extract element displacement vector %——— for i: eldisp(i)=disp(index(i)); end % compute strains % compute stresses % store for each element stress(intp,i)=estress(i); % store for each element end % location=fielp,intx jinty,intz] % print location for stress stress(intp,:) % print stress values end end end % end of integration loop % end % end of loop for total number of elements % % function [dhdx,dhdy,dhdz]=federiv3(nnel,dhdr,dhds,dhdt,invjacob) % % Purpose:Section 9.9 MATLAB Application to 3-D Stress Analysis 351 % determine derivatives of 3-D isoparametric shape functions with % respect to physical coordinate system % % Synopsi % [dhdx,dhdy,dhdz]=federiv3(nnel,dhdr,dhds,dhdt,invjacob) % % Variable Description: % dhdx - derivative of shape function w.r.t. physical coordinate x % dhdy - derivative of shape function w.r.t. physical coordinate y % dhdz - derivative of shape function w.r.t. physical coordinate z % nnel - number of nodes per element % dhdr - derivative of shape functions w.r.t. natural coordinate r % dhds - derivative of shape functions w.r.t. natural coordinate s % dhdt - derivative of shape functions w.r.t. natural coordinate t % invjacob - inverse of 3-D Jacobian matrix invjacob(1,1)*dhdr(i)-+invjacob(1,2)*dhds(i) ... +invjacob(1,3)*dhdt(i); dhdy (i)=invjacob(2,1)*dhdr(i)-+invjacob(2,2)*dhds(i) ... +inyjacob(2,3)*dhdt(i); dhdz(i)=invjacob(3,1)*dhdr(i)-+invjacob(3,2)*dhds(i) ... +tinvjacob(3,3)*dhdt(i); end % % function (shapes8 dhdrs8,dhdss8,dhdts8]=feisos8(rvalue,svalue,tvalue) % % Purpose: % compute isoparametric eight-node solid shape functions % and their derivatves at the selected (integration) point % in terms of the natural coordinate % % Synopsis: % [shapes dhdrs8,dhdss8,dhdts8]=feisos8 (rvalue,svalue,tvalue) w % Variable Description: % shapes8 - shape functions for four-node element % dhdrs8 - derivatives of the shape functions w.r.t. r % dhdss8 - derivatives of the shape functions w.r.t. s % dhdts8 - derivatives of the shape functions w. % rvalue - r coordinate value of the selected point % svalue - s coordinate value of the selected point % tvalue - t coordinate value of the selected point352 Elasticity Problem Chapter 9 % % Notes: % Ast node at (-1,-1,-1), 2nd node at (1,-1,-1) % 3rd node at (1,1,-1), 4th node at (-1,1,-1) % Sth node at (-1,-1,1), 6th node at (1,-1,1) % 7th node at (1,1,1), 8th node at (-1,1,1) %: % % shape functions .125*(1-rvalue)*(1-svalue)*(1-tvalue); .125*(1+rvalue)*(1-svalue)*(1-tvalue); .125*(1+rvalue)*(1+svalue)*(1-tvalue); 125*(1-rvalue)*(1-+svalue)*(1-tvalue); .125* (1-rvalue)*(1-svalue)*(1+tvalue); 125*(14rvalue)*(1-svalue)*(1-+tvalue); 125*(1+rvalue)*(1+svalue)*(1+tvalue); .125*(1-rvalue)*(1-+svalue)*(1-4+tvalue); % derivatives % dhdrs8(1)= 125*(1-svalue)*(1-tvalue); dhdrs8(2)=0.125*(1-svalue)*(1-tvalue); dhdrs8(3)=0.125*(1+svalue)*(1-tvalue); dhdrs8(4)=-0.1254(1+svalue)*(1-tvalue); dhdrs8 (5)=-0.125*(1-svalue)*(1+tvalue); dhdre8(6)=0.125*(1-svalue)*(1+tvalue); dhdrs8(7)=0.125*(1+svalue)*(1+tvalue); dhdrs8(8)=-0.125*(1+svalue)*(1+tvalue); % dhdss8(1)=-0.125*(1-rvalue)*(1-tvalue); dhdss8(2)=-0.125*(1+rvalue)*(1-tvalue); dhdss8(3)=0.125*(1+rvalue)*(1-tvalue); dhdss8(4)=0.125*(1-rvalue)*(1-tvalue); dhdss8(5)=-0.125*(1-rvalue)*(1+tvalue); dhdss8(6)=-0.125*(1+rvalue)* (1+tvalue); dhdss8(7)=0.125*(1+rvalue)*(1+tvalue); dhdss8(8)=0.125*(1-rvalue)*(1+tvalue); % dhdts8(1)=-0.125*(1-rvalue)*(1-svalue); dhdts8(2)=-0.125*(1+rvalue)*(1-svalue); dhdts8(3)=-0.125*(1+rvalue)*(1+svalue); dhdte8(4)=-0.125*(1-rvalue)*(1+svalue); dhdts8(5)=0.125*(1-rvalue)*(1-svalue); dhdts8 (6)=0.125*(1+4rvalue)* (1-svalue); dhdts8(7)=0.125*(1+rvalue)*(1-+svalue); dhdts8(8)=0.125*(1-rvalue)*(1+svalue); %Section 9.9 MATLAB Application to 3-D Stress Analysis % function [jacob3]=fejacob3(nnel,dhdr,dhds,dhdt,xcoord,ycoord,zcoord) % % Purpose: % determine the Jacobian for three-dimensional mapping % % Synopsis: % (jacob3}=fejacob3 (nnel,dhdr,dhds,dhdt, xcoord,ycoord,zcoord) % % Variable Description: % jacob3 - Jacobian for one-dimension % nnel - number of nodes per element % dhdr - derivative of shape functions w.r.t. natural coordinate r % dhds - derivative of shape functions w.r.t. natural coordinate s % dhdt - derivative of shape functions w.r.t. natural coordinate t % xcoord - x axis coordinate values of nodes % ycoord - y axis coordinate values of nodes % acoord - 2 axis coordinate values of nodes % % jacob3=zeros(3,3); % for i=I:nnel jacob3(1,1)=jacob3(1,1)+dhdr(i)*xcoord(i); jacob3(1,2)=jacob3(1,2)+dhdr(i)*ycoord(i); jacob3(1,3)=jacob3(1,3)-+dhdr(i)*zcoord(i); jacob3(2,1)=jacob3(2,1)-+dhds(i)*xcoord(i); jacob3(2,2)=jacob3(2,2)+dhds(i)*ycoord(i); jacob3(2,3)=jacob3(2,3)-+dhds(i)*zcoord(i); jacob3(3,1)=jacob3(3,1)-+dhdt(i)*xcoord(i jacob3(3,2)=jacob3(3,2)+dhdt(i)*ycoord(i jacob3(3,3)=jacob3(3,3)-+dhdt(i)*zcoord(i); end % % function [kinmtx3]=fekine3d(nnel,dhdx,dhdy,dhdz) a % Purpose: % determine the kinematic equation between strains and displacements % for three-dimensional solids % ‘% Synopsis:Elasticity Problem Chapter 9 % (kinmtx3]=fekine3d (nnel,dhdx,dhdy,dhdz) % % Variable Description: % nnel - number of nodes per element % dhdx - derivatives of shape functions with respect to x % dhdy - derivatives of shape functions with respect to y % dhdz - derivatives of shape functions with respect to z kinmtx3(1,i1 kinmtx3(2,i2 kinmtx3(3,i3)=dhdz(i); kinmtx3(4,i1)=dhdy (i); kinmtx3(5,i2 kinmtx3(5,i3 kinmtx3(6,i1 kinmtx3(6,i3 end % %— Nodal displacements are given below and the state of stresses is 7, = 1000 and the rest of stresses are zero at every integration point. displace = dof. displ. 1.0000 0.0000 % x-displacement of node 1 2.0000 0.0000 % y-displacement of node 1 3.0000 0.0000 % z-displacement of node 1 4.0000 -0.0030 % x-displacement of node 2 5.0000 0.0000 % y-displacement of node 2 6.0000 0.0000 % z-displacement of node 2 7.0000 -0.0030 % x-displacement of node 3 8.0000 -0.0030 % y-displacement of node 3 9.0000 0.0000 % 2-displacement of node 3 10.000 0.0000 % x-displacement of node 4 11.000 -0.0030 % y-displacement of node 4 12.000 0.0000 % z-displacement of node 4 13.000 0.0000 % x-displacement of node 5 14.000 0.0000 % y-displacement of node 5 15.000 0.0100 % w-displacement of node 5 16.000 —-0.0030 % x-displacement of node 6 17.000 0,000 % y-displacement of node 6Section 9.9 18.000 19.000 20.000 21.000 22.000 23.000 24.000 MATLAB Application to 3-D Stress Analysis 0.0100 0.0030 .0030 0.0100 0.0000 0.0030 0.0100 % z-displacement of node 6 % x-displacement of node 7 % y-displacement of node 7 % z-displacement of node 7 % x-displacement of node 8 % y-displacement of node 8 % z-displacement of node 8 355356 91 9.2 9.3 94 Elasticity Problem Chapter 9 Problems An orthotropic material under the plane stress condition has the constitutive equation {7} = [D}{e} where {@} and {@} are the stress and strain vectors in terms of the material orthotropic directions. Further, [D] is given as . Qu Q2 0 [D]=|Q%2 Q2 0 0 0 Oss where oe 1~ raver Qu rnaEe near En cle peer 1-201 Qs3 = Giz Subscripts 1 and 2 denote the orthotropic axes, E, and E» are elastic modulii in the 1 and 2-axis, and Gy is the inplane shear modulus. 14; is Poisson’s ratio for the normal strain ¢; resulting from the normal strain ¢. If the material orthotropic axes are rotated from the global coordinate axes by angle 8 (see Fig. P9.1), show that the strain vectors between the two coordinate systems are related as shown below: 41 & ists) ‘2 Yey Qn= Q: where cos? sin? cosOsind (T= sin?6 cos?0 —cos0sind —2cos8sind 2cosfsin0 cos?6 — sin?6 From Prob. 9.1, also prove that [D] = [7]*[D][T] where [D] is the material property matrix in terms of z — y coordinate system and [D] is the material property matrix in terms of 1 — 2 coordinate system. A two-dimensional elastic body is discretized using six-node triangular elements as shown in Fig. P9.3. Find the equivalent nodal forces for the three nodes located on an element boundary. Find the equivalent nodal forces for Fig. P9.4.Problems 357 2 Figure P9.1 Problem 9.1 Tp Les, , Figure P9.3. Problem 9.3 9.5 9.6 9.7 98 9.9 9.10 Find the equivalent nodal forces for Fig. P9.5. Find the deflection of a tapered cantilever beam as shown in Fig. P9.6 using linear triangular elements and the computer programs, Assume unit width of the beam. Solve Fig. P9.7 using computer programs. Use different mesh discretization. Some elements are distorted as shown in the figure. Modify the computer programs so that linear triangular elements can be used with rectangular elements. Then solve Fig. P9.8. Using the computer programs, find the solution for a perforated plate under tension as seen in Fig. P9.9. Solve Prob. 9.6 for the transient analysis assuming the beam is initially at rest using the computer programs.358 Elasticity Problem Chapter 9 20kN/m O.im 0.1m O.1m Qin Figure P9.4 Problem 9.4 40KN/m Dim) Dim) Om | Ol Figure P9.5 Problem 9.5 peel E=2106P 0, Figure P9.6 Problem 9.6Problems 359 lcm | 8cm 200N / Ea 10cm — 4000 / 4000 fen, 10cn, Ben teen £27560 Figure P9.7 Problem 9.7 a E=200GPa 0.3m 3kN 0.3m 0.3m O.3m_|_ 0.3m | 0.3m Figure P9.8 Problem 9.8 0.4m by 0.3m plate with 0.05m radius hole at the center symmetry SkN/m symmetry E=200GPa Figure P9.9 Problem 9.9CHAPTER TEN PLATE STRUCTURES 10.1 Classical Plate Theory The basic assumptions for the classical Kirchhoff plate bending theory are very similar to those for the Euler-Bernoulli beam theory. One of the most important assumptions for both theories is that a straight line normal to the midplane of the plate (or beam) before deformation remains normal even after deformation. In other words, the transverse deformation is neglected. Therefore, as shown in Fig. 8.1.2, inplane displacements u and v can be expressed as us-25 (10.1.1) dw vee (10.1.2) where x and y are the inplane axes located at the midplane of the plate, and z is along the plate thickness direction as seen Fig. 10.1.1. In addition, u and v are the displacements in the x- and y-axes, respectively, while w is the transverse displacement (or called deflection) along the 2-axis. Because we neglect the transverse shear deformation, inplane strains can be written as in terms of the displacements fe y= Che eo) (10.1.3) where : : : {nm}? ={ee hy yl = (FS Fe 2H} (10.1.4) is called curvature. Assuming the plane stress condition for plate bending and substituting Eqs (10.1.3) and (10.1.4) into Eq. (9.1.5) yield the constitutive equation as given below: {o} = -2[D]{x} (10.1.5) 361362 Plate Structures Chapter 10 Figure 10.1.1 Free Body Diagram of the Plate Element in which {c} = {oz oy Tey }” and [D] is defined in Eq. (9.1.6). Moments are defined as hfe {M} =| {o}z dz (10.1.6) nye where {M} ={M, M, Mzy} and his the plate thickness. Substitution of Eq. (10.1.5) into Eq. (10.1.6) gives the relationship between the moment and curvature. {M} = -[Dl{x} (10.1.7) where it 1B) = FI (10.1.8) Equilibrium equations are obtained from the free body diagram as shown in Fig. 10.1.1. Moment equilibriums about y- and 2-axes and force equilibrium about z-axis yield after neglecting higher order terms OM, , OMey _ Fa + Heh Qe =0 (10.1.9) OMay , OM, eee i eeeeeme = Gat + yi Ww =0 (10.1.10)Section 10.1 Classical Plate Theory 363 80 , OQy ae” Oy where Qe and Qy are the shear forces and p is the distributed pressure loading as seen in Fig. 10.1.1. Elimination of the shear forces from Eqs (10.1.9) through (10.1.11) gives +p=0 (10.1.1) OM, , 0?Mey , ?M, Oa? + by + ay +p=0 (10.1.12) Combining Eqs (10.1.4), (10.1.7) and (10.1.12) finally produces the biharmonic governing equation for plate bending in terms of the transverse displacement w. Ow otw dw op bat + Batay * By = D, (10.1.18) where D, = aM5zy is the plate rigidity. 4 Example 10.1.1 We want to detive the equilibrium equations, Eqs (10.1.9) through (10.1.11), from Eqs (9.4.1) through (9.4.3). Integration of Eqs (9.4.1) and (9.4.2) over the plate thickness after multiplying them by z yields 0% , Otay , ter) [Gee Bee + Ge ede = OM, , 0Mzy a2 + By = Qe + [resel ig = 0 (10.1.14) [°0 (Set Oe Pe) ete =h/2 ae * dy 8: OMry , OM, epee. a; SY — Qy + [tyz2]” ee (10.1.15) in which a] a= f Tey (10.1.16) —h/2 na Q,= [ Tyde (10.1.17) =h/2 If there is no shear stress (Tey = Ty: = 0) on the top and bottom surfaces of the plate, Eqs (10.1.14) and (10.1.15) are equal to Eqs (10.1.9) and (10.1.10). Integrating Eq. (9.4.3) over the plate thickness gives M3 (Ore, Dies , 802 )= = + dz Lol oy toe we +e 4+.0,(h/2) — 0;(—h/2) = (10.1.18)364 Plate Structures Chapter 10 UE (ep.vp) x (40%) Figure 10.2.1 Three-Node Plate Bending Element If o,(h/2) = p and o,(—h/2) = 0, Eq. (101.18) equals to Eq. (10.111). + 10.2 Classical Plate Bending Element We derive a three-node plate bending element based on the classical plate theory [34]. The clement is shown in Fig, 10.2.1. Each node of the element possesses three degrees of freedom: displacement w in the z direction; a rotation about the z-axis, wy (derivative of w with respect to y); and a rotation about the y-axis, we (derivative of w with respect to z), as shown in the figure. The displacement function w is assumed to be w(2,y) = a +a22+asy+agz? +aszyt acy? + a72° + ag(2?y + zy”) +a9y° = [X]{a} (10.2.1) where [X]J=[1 2 y 2? ay y? 2% (22y+ey) y?] (10.2.2) {a}={a, az a3 ay as a5 a7 as}” (10.2.3) Here constants a; are to be replaced by nodal variables w, wz and wy. Taking derivatives of the displacement function with respect to z and y yields as = ay + 2age + agy + Bare? + ag(22y + y?) (10.2.4) bw : : Gp 0+ 88 + Pay + ase? + Day) + Saoy (10.2.5)Section 10.2 Classical Plate Bending Element 365 Evaluation of Eqs (10.2.1) through (10.2.5) at the three nodal points gives the following matrix expression: {a} = [X]{a} (10.2.6) where {d}={wr (we): (wy): we (we)2 (wy)2 ws (we)s (wy)s}7 (10.2.7) and a om 2} am vf tf cinta, of 10 24) mw 0 3827 22imt+y? 0 0: 12..0 w 2m 0 xi+2e1y, Sy? 22 U2 23 2th yy 23 thy t+ coyZ 1 0 0 1 5 [X]=]0 1 0 22 yo 0 322 2ayty? 0 (20.2.8) 0 7 0 0 Ot 0 ay os 0 a Se 23 vs 23 toys uf 23 xiustzsus ue 1 0 2% ys 0 322 2zays+ue 0 0 1 0 23 2% 0 2242243 3y2 Inverting Eq. (10.2.6) and substituting the result into Eq. (10.2.1) yields {w} = [H}{a} (10.2.9) where the row vector of shape functions of size 9 x 1 is computed from (H) = [X]xT-* (10.2.10) Inplane strains are computed from Eq. (10.2.9) as {e} = [B]{a} (10.2.1) in which {h={eo cy rey)” (10.2.12) [B) = - (0X? (10.2.13) 000200 32 0 w=[p 000020 2% by (10.2.14) 000020 0 4e+y) 0 Substitution of the strain-nodal displacement relation, Eq. (10.2.11), into the expression for element stiffness matrix, Eq. (9.1.24) yields wl= f ftertoneys an ater? ff tutta an (x = [x7 f or pinanrsy (10.2.15)366 Plate Structures Chapter 10 where eh = Fl) (10.2.16) Here (D] is the constitutive matrix of the plane stress condition, Q¢ is the two- dimensional element domain in the zy-plane, and h is the thickness of the plate. The element domain is the triangular shape as seen in Fig. 10.2.1. The element stiffness matrix is of size 9 x 9 and the corresponding element nodal vector is given in Eq. (10.2.7). 10.3 Shear Deformable Plate Element The Mindlin/Reissner plate theory includes the effect of transverse shear defor- mation like the Timoshenko beam theory. Hence, a plane normal to the midplane of the plate before deformation does not remain normal to the midplane any longer after deformation, The internal energy expression for the shear deformable plate should include transverse shear energy as well as bending energy. The internal energy is expressed as u 5 [tor tonay +5 [ to fenav (10.3.1) where {o}={or oy tey}" (10.3.2) fa}={ee ey tev)” (10.3.3) are the bending stress and strain components while {o.}= {te ty)" (10.3.4) fe} = {722 ys)" (10.3.5) are the transverse shear components. In addition, « is the shear energy correction factor and equal to 8. Substitution of the constitutive equations for both bending and shear compo- nents yields v=} i {a} Dila)av +5 [ {e}7(D.He}av (10.3.6) in which 0 0 | (10.3.7) = 3 (10.3.8)Section 10.3 Shear Deformable Plate Element 367 Further, V is the three-dimensional domain which is equal to dQ x dz. The zyz coordinate axis is the same as shown in Fig. 10.1.1 In order to derive the element stiffness matrix for the shear deformable plate, we need to express the strains in terms of nodal variables. The inplane displacements are given as u=—20,(2,y) (10.3.9) v= —z6y(z,y) (10.3.1) and the transverse displacement is w= w(z,y) (10.38.11) where 0, and 0, are rotations of the midplane about y and z axes, respectively. The midplane is assumed to have no inplane deformation. For the shear deformable plate, a 6,= oa — Yee (10.3.12) dw 85 = Fy 70 (10.3.18) where 7 is the angle caused by the transverse shear deformation as seen in Fig. 8.2.1. Because the transverse displacement w and slope 0 are independent, we need shape functions to interpolate them independently. As a result, the shear deformable plate element requires C° compatibility. Isoparametric shape functions are used for the plate clement formulation. The transverse displacement and slopes are interpolated as w= De, nw (10.3.14) ist 6,5 Sime, (92) (10.3.15) isl 8y = YO HilEsn) (Oy) (10.3.16) is Here n is the number of nodes per element and the same shape functions are used for the displacement and slope interpolations. For the following presentation, bilinear isoparametric shape functions are used for simplicity. However, higher-order shape functions can be used in the same manner. Both bending and shear strains are computed from displacements. {eo} = -2[Bsl{d"} (10.3.17) {es} = [Bsl{a*} (10.3.18)368 Plate Structures Chapter 10 where a0 0 0 eo eo 0 l={ 9 4 0 0 % 0 0 eo 0 FF 0] (10.3.19) on, oh 9 oH, oh 9 ams a, 9 Oe othe 9 32 oy oe oy Or ay be {B.] = an 0 4 -H, 0 4 -wy 0 4 -H, 0 i a 0 -H, 4 0 -H, 4 0 -H, 4 0 -H, 4H (10.3.20) and {d°}= {(2): (Oy)x wi (Ox)2 (Oy)2 we (Ox)s (Ay)a wa (x)s (Oy) wa)” (10.3.21) Substitution of Eqs (10.3.17) and (10.3.18) into the energy expression Eq. (10.3.6) yields for each plate element oa say [fer oawadanays Sey [fe wae.ldcanter) (10.3.2) As a result, the element stiffness matrix for plate bending can be expressed as (K*] & i (Ba (DaBu}d0 + wh f (BAF (DllB.la0 (10.3.23) in which h is the plate thickness. One thing to be noted here is that the shear energy becomes dominant compared to the bending energy as the plate thickness becomes very small compared to its side length. This is called shear locking. A heuristic explanation for this can be given as below. The bending energy is proportional to h® while the shear energy is propotional to h. Therefore, as b gets smaller, the shear energy becomes dominant over the bending energy. To resolve this problem, the selective or reduced integration technique was proposed. The key of the technique is to underintegrate the shear energy term. In general, the bending term is integrated using the exact integration rule. For example, when four-node bilinear isoparametric elements are used, the 2 x 2 Gauss-Legendre quadrature is used for the bending term while I-point integration is used for the shear term. Similarly, the bending term utilizes 3 x 3 integrations and the shear term uses 2 x 2 integrations for the nine-node biquadratic isoparametric shape function. 10.4 Plate Element With Displacement Degrees of Freedom The plate bending element developed in this section is shown in Fig. 10.4.1 where z, y, and z describe the global coordinate of the plate and u, v, and w are theSection 10.4 Plate Element With Displacement Degrees of Freedom 369 au Figure 10.4.1 Plate Element With Displacement Degrees of Freedom displacements. h is the plate thickness. The zy plane is parallel to the midsurface plane prior to deflection. The displacement of any point in the plate can be expressed as u=u(z,y,2) (10.4.1) v=v(z,y,2) (10.4.2) w=u(z,y) (10.4.3) ‘That is, the inplane displacements u and v vary through the plate thickness as well as within the zy-plane while the transverse displacement w remains constant through the plate thickness [35,36]. In order to interpolate the displacements using shape functions and nodal displacements, two different interpolations are needed: one interpolation within the zy-plane and the other in the z-axis. For the zy-plane interpolation, shape functions N;(z, y) ate used where subscript i varies depending on the number of nodes on the zy-plane. On the other hand, shape functions H(z) are used for interpolation along the z-axis, where subscript j varies depending on the number of nodes along the plate thickness. Because two inplane displacements are functions of z, y, and z, both shape functions are used while the transverse displacement uses shape functions Ni(e,y). Using isoparametric elements with mapping of £7-plane onto zy-plane and ¢-axis to z-axis, the three displacements can be expressed as Ni Na ws DD ME mH (ais (10.4.4) i=l j=l Ms Na v= DOD ME mH (Cvs (10.4.5) i= j=l MM w= SNE nui (10.4.6) a370 Plate Structures Chapter 10 in which N; and No are the numbers of nodes in the zy-plane (7-plane) and z-axis (¢-axis), respectively. In addition, the first subscript for u and v denotes the node numbering in terms of zy-plane (Eq-plane) and the second subscript indicates the node numbering in terms of z-axis (¢-axis). In the present study, N,=4 and N2=2. That is, four-node quadrilateral shape functions are employed for the zy-plane (€7- plane) interpolation and linear shape functions are employed for the z-axis (¢-axis) interpolation. Nodal displacements u;; and v;; are displacements on the bottom surface of the plate element and uj2 and vj2 are displacements on the top surface. As seen in Eqs (10.4.4) through (10.4.6), there is no rotational degree of freedom for the present plate bending element. In the present formulation, both bending strain energy and transverse shear strain energy are included. The bending strains and transverse shear strains are expressed in terms of displacements. &e je 0 07 wef y |. F y | { | (10.4.7) You & & OL lw ta)={%} = % z| {3} (10.4.8) where {¢s} is the bending strains and {¢,} is the transverse shear strains. The normal strain along the plate thickness ¢, is omitted here. Substitution of displacements, Eqs (10.4.4) through (10.4.6), into the kinematic equations, Eqs (10.4.7) and (10.4.8), with Ny=4 and N2=2 expresses both bending and shear strains in the following way. e tho {es} = [Bi}{a"} (10.4.9) where (Bs] = [[Bn] [Bs] [Bs] [Bs4]] (0.4.10) Ae 0 eee 0 0 (=| 0 oe ee (10.4.1) HR He WM WMO. {a} = {4} {48} {di} {45} )7 (10.4.12) {dj}={ua v1 ui2 v2 wi} (10.4.13) {es} = [Bi]{d*} (10.4.14) where (B.]=((8.1] [Bex] [Bes] [Bsa] (10.4.15) Nob 0 Ng [Bui] i Nth ? i a. (10.4.16)Section 10.4 Plate Element With Displacement Degrees of Freedom. 371 The constitutive equation for the isotropic material is {oo} = [Dsl{es} (10.4.17) where {o}={o2 oy Tey} (10.4.18) [Di] = (10.4.19) for the bending components and {os} = [De}es} (10.4.20) where {os}={tyz tex 3” (10.4.21) E_f1 0 [DJ = a+) [3 1 (10.4.22) where Eq. (10.4.19) is the material property matrix for the plane stress condition as usually assumed for the plate bending theory. For a unidirectional fibrous composite, the material property matrices become Di Dp 0 [Di)=|Di2 Dn 0 (10.4.23) 0 0 Des in which (10.4.24) (10.4.25) (10.4.26) (10.4.27) and -[G@s 0 =[ — (10.4.28) Here, 1 and 2 denote the longitudinal and transverse directions of the undirectional composite, respectively. Further £ is the elastic modulus, Gj; is the shear modulus of the i— j plane and 14; is Poisson’s ratio for strain in the j-direction when stressed in the i-direction. There are five independent material properties for Eqs (10.4.23) through (10.4.28) because of the reciprocal relation 42 = 434.372 Plate Structures Chapter 10 The total potential energy can be expressed as N=U-w (10.4.29) where the internal strain energy U consists of two parts like U=U+U, (10.4.30) The bending strain energy Us is nes f {os}? {es}a0 (10.431) and the transverse shear strain energy is u, = 5 [ont (earan (10.4.32) where 91 is the plate domain. After finite element discretization, substitution of the previous equations into Eqs (10.4.31) and (10.4.32) gives w= Fey? f teat alssanter) (20.4.33) w= Dey f wt water) (10.434) where summation is performed over the total number of finite elements and superscript e indicates each element. Kinematic matrices [By] and [B,] are provided in Eqs (10.4.10) and (10.4.15) while the constitutive matrices [Dj] and [D,] are given in Eqs (10.4.19) and (10.4.22) for the isotropic material, and in Eqs (10.4.23) and (10.4.28) for the unidirectional composite. For a laminated composite plate, the material property matrix of each layer must be transformed based on the fiber axis of each layer and the global reference coordinate system. The external work is written as W = {d}"{F} (10.4.35) in which {d} is the system nodal displacement vector and {F°} is the system force vector. Because there is no rotational degree of freedom for the present element, the external moment is applied to the force vector as a couple applied on the top and bottom nodes of the plate element as shown in Fig. 10.4.1. Finally invoking the stationary value of the total potential energy yields the finite element matrix equation. The element stiffness matrix can be expressed as [kK] = if [Bal [Da] Be]d0 + f [Bi (Di 1B.1a0 (10.4.36)Section 10.5 Mixed Plate Element 373 One thing to be noted here is that the transverse shear strain energy term should be under-integrated numerically to avoid shear locking, especially for a thin plate. 10.5 Mixed Plate Element The basic equations for the classical plate theory are ew, ow Mz = -0.(53 + 5a) (10.5.1) 2, 2 = -».(5¢ +053) (10.5.2) May = —D-(1— nee (10.5.3) @M, | OM, , 8M, eeucccecees eee ee tat t oy + ag =P (10.5.4) where M is the moment and D, = wy is the flexural rigidity of plate. E is the elastic modulus, h is the plate thickness, v is Poisson’s ratio, and p is the pressure loading. Equations (10.5.1) through (10.5.3) are the constitutive equations and Eq. (10.5.4) is the equilibrium equation of moments. Applying Galerkin’s method to Eqs (10.5.1) through (10.5.4) do not produce the symmetric matrix. To this end, Eqs (10.5.1) through (10.5.3) are inverted so that we have S(Mz — vMy) + = =0 (10.5.8) S(My — vMz) +2 =0 (10.5.6) 25(1+v)Mey + 2 ow 5 =0 (10.5.7) where $ = 225. Now Galerkin’s method is applied to Eqs (10.5.4) through (10.5.7) and integration by parts are performed to develop the weak formulation. The resultant matrix equation for each element is given below [21]: ‘Ki Aa 0 Aa Me FR Ky K 0 KlJM |_Jh% 0 0 Ks Ke|) Myf) Fs oO ky Ka Ke 0 w Ki where fee f[ [NTF [Neo (10.5.9) le374 Plate Structures Chapter 10 Ky=-vKy (10.5.10) Kg = -[ anya a (10.5.11) Ks =21+v)Ki (10.5.13) vA (ETE @ ee =-f rst, av (10.5.15) R= =f wr Sy ar (10.5.16) ~ ft (Gre + Feu Jar (10.5.17) 7 [oF onar + f. A) (10.5.18) Qn = Qele + Qyly (10.5.19) Here, l; and ly ate direction cosines of the unit normal vector, and Q is the shear force. [N] is the shape function vector. Any isoparametric element, of either quadrilateral or triangular shape, may be used for these equations. However, the previous formulation does not include the effect of transverse shear deformation. The mixed plate bending formulation for thick plates is derived below. Equilibrium equations for plate can be written as below including transverse shear forces. OM, , OMry oy (10.5.20) OMay , OM, Saree eee ce = da tay = 0 (10.5.2) 202 , 2p =0 (10.5.22) az Oy The major discrepancy between the thin and thick plate theories is the relations between the rotations and the transverse deflection. In the thin plate theory the rotations are not independent of the transverse deflection but they are independent of the deflection for the thick plate theory. Thus, the displacements in the z, y and z directions are expressed as u=—z6,(z,y) (10.5.23) v = —26y(z,y) (10.5.24)Section 10.5 Mixed Plate Element. 375 w= w(z,y) (10.5.25) where 6, and y are rotations about y and z axes, respectively. Substitution of Eqs (10.5.23) through (10.5.25) into the kinematic equations and use of the constitutive equations give S(Me—vMy) + oe =0 (10.5.26) - ED S(My — Mz) + GY = 0 (10.5.27) 6, 28(1 +») Mey + oe +h =0 (10.5.28) If 6, and 6y are replaced by $% and $, Eqs (10.5.26) through (10.5.28) are the same as Eqs (10.5.5) through (10.5.7). Such relations, however, do not hold in the thick plate theory. Using the constituent and kinematic equations for transverse shear components, the shear forces can be expressed in terms of rotations and the deflection. dw Qe = nGH(—05 + 5) (10.5.29) a Qy = nGh(—y + ra (10.5.30) where « is the shear correction factor equal to 5/6, G is the shear modulus, and h is the plate thickness. Rewritting Eqs (10.5.29) and (10.5.30) for the rotations yields —_Q: , dw tg ech 4 02 (105.31) es. Ow 6 =-t a (10.5.32) Putting Eqs (10.5.31) and (10.5.32) into Eqs (10.5.26) through (10.5.27) to eliminate the rotations gives 1 OQ, , Ow S(Mz - vMy) ~ ae SE + SG = 0 (10.5.3) 1.00, & S(My ~My) = 2, 8d 4 FS = (10.5.34) i/o. 6) ew 25(1 +0) May - al} 7 %) +25 (10.5.5) Examining Eqs (10.5.33) through (10.5.35) reveals that the coefficients of shear forces and moments are of order 1/h and 1/h°, respectively. Thus, as the plate thickness approaches zero, the shear force terms can be neglected compared to the moment. terms. This is reasonable because the shear deformation is negligible when the plate thickness is very small compared to its length.376 Plate Structures Chapter 10 In order to eliminate the shear forces, Eqs (10.5.20) and (10.5.21) are substituted into Eqs (10.5.33) through (10.5.35) as well as Eq. (10.5.22). Then, the resultant equations are 1 oe 1 O'My , &w (5~ sande) Mo 95M - eg wah Beiy * Fa? 79 (10.5.36) 13 1 PMgy , Ow Bae _-— — Saas ees ee vSMz + (s aig) Me Gh Bede + Hz =9 (10.5.37) 1 (8°M, OM, ‘My 1 @ 1 @ aw ~ san (Seas + Saat) + (28049) ~ sepa wan byt) Mev +? 12 y = (10.5.38) as well as Eq. (10.5.4). Equations (10.5.36) through (10.5.38) and (10.5.4) have the same four variables, M,, My, Mcy and w, as those for the thin plate formulation. If the terms associated with =, are neglected, these equations are reduced to the thin plate equations. In fact, as the plate thickness approaches zero, these terms are neglected. For the shear related terms are proportional to 4 while bending related terms are proportional to js. Applying Galerkin’s method to the four equations yields the following matrix expression. Kn Ki Kis Kia Mz A Kin Ko. Koy Kaa M \_J% Kis Koa Ks3 Kaa} ) Moy [~ ) Fs oo) hig Kog Kag Kaa w Fy where Ky=S i INP wan + a f ey el" (10.5.40) a, [NF [yaa (10.5.41) kus= op fe wy [= “i (10.5.42) Ku=- Liz a [ay (10.5.43) los an iT Seca Kn=5 iy wlan + bs f (10.5.44) Tene (10.5.45) 28 = TGR Ine S s ON ON ee i: eae sy aa (10.5.46)Section 10.5 Mixed Plate Element 377 4 Elements 16 Buments 7 ' { 1 L L > allel ' | i i | i t Tt i ay a ae aed Figure 10.5.1 Square Plate Mesh Using 4 and 16 Elements : riven s 2 f ONT" [ON Ko =a+0)s f wwrimans yf [FE] [BE] e+ a l. any" ge dQ (10.5.47) “Au (Ce] Eee] + [Se] [Sp] ose Ku =0 (10.5.49) — | [vy" bar + [NJ]? VeledP (10.5.50) ow ener + ef. ~ for nar + Ze | wr vote (10.5.51) =-f. iw F (Fete + Fi Sey jars wer wr (F (Mele + Mey + (Mayle + My ))ar (10.5.52) = i [IN]? Qndl + f LN] pate (10.5.53) i Ke Some finite element solutions obtained using the present mixed plate bending elements are shown in Tables 10.5.1 through 10.5.5. Isoparametric shape functions were used for both interpolation of moments and displacements. Tables 10.5.1 and 10.5.2 show the results for simply supported and clamped square plates subjected to uniform pressure loads. Because of symmetry, 4 or 16 four-node isoparametric elements were used. The finite element mesh is seen in Fig. 10.5.1. The solutions from the present mixed formulation are also compared with those from another mixed formulation. Table 10.5.3 gives the finite element solutions obtained using 4 eight- node isoparametric elements while Table 10.5.4 compares different isoparametric plate bending elements. The accuracy of each isoparametric element is different even if the378 Plate Structures Chapter 10 Table 10.5.1 Comparison of Central Deflections and Bending Moments for a Uniformly Loaded Simply Supported Square Plate. Analytic | 4 Elem.* | 16 Elem.” | 4 Elem.*" | 16 Elem."* Solution. WD/PL* | 0.00406 0.00424 0.00411 0.00409 0.00407 M./PL? 0.0479 0.0525 0.0489 0.0505, 0.0485 M,/PL? 0.0479 0.0525 0.0489 0.0505, 0.0485 4-node quadrilateral element (*) - Present F.E. Solution, (**) Solution from [20] ‘Table 10.5.2 Comparison of Central Deffections and Bending Moments for a Uniformly Loaded Clamped Square Plate. Analytic | 4 Elem." | 16 Elem.” | 4 Elem.** | 16 Elem.** Solution. WD/PL4 | 0.00126 0.00141 0.00128 0.00148 0.00132 M,/PI? -0.0513 -0.0476 -0.0499 -0.0487 -0.0508 M,/PL? -0.0513 -0.0476 -0.0499 ~0.0487 ~0.0508 4-node quadrilateral element (*) - Present F.E. Solution, (**) Solution from [20] total numbers of nodes are almost the same. The elements with more nodes per element give more accurate results. Both the thin plate theory and the thick plate theory are compared in Table 6.5.5 for an orthotropic plate. The plate is shown in Fig. 6.5.2 with the mesh and material properties. As expected, as the ratio of the plate thickness to the side length increases, there is an increasing difference between the two solutions. The thick plate solutions are very close to the three-dimensional elasticity solutions for thick plates. 10.6 Hybrid Plate Element The hybrid element is based on the assumed strains within the plate element. [23]. This element requires C° continuity. The formulation is based on a modified potential energy expression as given below for a plate. a= f (- Hal" Dalle) — jo OMe) + (oF Dalla aSection 10.6 Hybrid Plate Element 379 Table 10.5.3 Deflections at a Center of Square Plate Boundary Condition Analytic Soln | Present Soln All Edges Supported 0.00406 0.00406 All Edges Clamped 0.00126 0.00125 ‘Two Opposite Edges Simply Supported Two Other Edges Clamped 0.00191 0.00192 8-node quadrilateral element Table 10.5.4 Comparison of Central Deflections Obtained Using Different Isoparametric Elements for Uniformly Loaded Simply Supported Plates We Error (%) Remark Analytic Solution 0.2363 Timoshenko 3-Node Triangular 0.1814 -22.81 32 Elements (25 Nodes) 6-Node Triangular 0.2344 -0.80 8 Elements (25 Nodes) 4-Node Quadrilateral 0.2392 1.23 16 Elements (25 Nodes) 8-Node Quadrilateral 0.2365, 0.08 4 Elements (21 Nodes) {e)7DalsHaan— [ (ay? pyar (106.1) where fo}=(% (M4 9}" (10.6.2) fea} = {(-0 + 98) (Oy + $8) }" (10.6.3) {d}={02 6, wi" (10.6.4) Further [Ds] is the material property matrix for bending strains and [D,] is the matrix for transverse shear strains. [Zs] is the matrix for the bending strain-displacement operator and [L,] is the matrix for the shear strain-displacement operator. {p} is the pressure loading on the plate.380 Plate Structures Chapter 10 Orthotropie Porperties Egy = 0.23319*E Ey = 0.543103*Ex Gry = 0.262931 *E x, Gog = 0.159914*E x, Cyz = 0.26681*E x Figure 10.5.2 Orthotropic Plate Invoking a stationary value of the equation results in the equilibrium equation and the generalized strain-displacement relation. In order to obtain the finite element model, generalized strains and displacements are discretized as the following: {es} = [B]{ao} (10.6.5) {es} = [Bsl{as} (10.6.6) {d} = [N]{d} (10.6.7) where generalized strains are assumed independently within each element and gener- alized displacements are interpolated using generalized nodal displacement {d}. Thus, [By] and [B,] are matrices consisting of the polynomial terms of the generalized strain parameter vectors {ap} and {a5}, respectively. [N] is the matrix consisting of shape functions. Substituting Eq. (10.6.5) through Eq. (10.6.7) into Eq. (10.6.1) yields == Flas) "IGiltas} — Feu} IGsHas) + {oo}? He} + {as} [Ha} ~ {d}7{F} (10.6.8) where ton = f waPtotenen (106.9) tel= f (wo. Ke.ua (10.6.10) feSection 10.6 Hybrid Plate Element 381 Table 10.5.5 Generalized Central Deflections (EzW/Pt) for Thin or Thick Uniformly Loaded Simply Supported Square Plates 3-D Reissner’s | Classical | Thin Plate | Thick Plate b/a | t/a | Theory* | Theory* | Theory* Soln** Soln** 0.05 21542 21542 21201 21268 21606 0.5 | 01 1408.5 1408.4 1325.1 1329.3 1413.8 0.14 | 387.23 387.27 344.93 346.03 389.11 0.05 | 10443 10442 | 10246 10285 10483 10} 01 688.57 688.37 640.39 642.81 692.30 0.14 191.07 191.02 166.70 167.33 192.49 0.05 | 2048.2 2047.9 1988.1 1964.6 2026.8 20 | 01 139.08 138.93 124.26 122.79 138.26 0.14 | 39.790 39.753 32.345, 31.962 39.806 4-node quadrilateral element (*) Analytical Solution, (**) Present F.E. Solution [i] = i ABs (Dal(Eatv]an (10.6.11) [H.) = iL [e Daltamian (10.6.12) and {F}= [ (N]" {p}ar (10.6.13) re Invoking stationary values of Eq. (10.6.8) with respect to {as} and {a} respectively results in —[Gs]{on} + [Ha]{d} = 0 (10.6.14) -[G.}{ae} + [H]{d} = 0 (10.6.15) Eliminating {op} and {a} from Eq. (10.6.8) using Eq (10.6.14) and Eq. (10.6.1) gives Une : : T= 3 (G17 LH) + (H.)(G.)-7H.){d} — {a}7{F} — (10.6.16) Equation (10.6.16) finally gives the following finite element system of equations [KI{a} = {F} (106.17)382 Plate Structures Chapter 10 Ss NORM. DEFLECTION aT NORM. DEFLECTION 3 3 oS 2x2 4x4 6x6 2x2 4x4 6x6 MESH IN A QUARTER PLATE MESH IN A QUARTER PLATE (@) (b) Figure 10.6.1 Uniformly Loaded Square Plate With: (a) Simply Support Boundary and (b) Clamped Boundary in which (K] = (Hel? (Gi? ie) + HIG.) (10.6.18) For a bilinear plate element, the generalized strain vectors are assumed as 100200400 [Bl=/0100200y0 (10.6.19) oe ee and wa=[} | (10.6.20) ‘These expressions represent that the bending strain varies linearly and the shear strain is constant within the bilinear plate element. Finite element results from the hybrid plate bending elements are provided in Fig. 10.6.1 through 10.6.3. Convergence study for simply supported and clamped square plates subjected to uniform pressure loading is shown in Fig. 10.6.1 while that for a uniformly loaded circular plate with the clamped edge is seen in Fig. 10.6.2. The mesh for the circular plate is shown in Fig. 10.6.3. 10.7 MATLAB Application A static finite element analysis of plate bending is performed using the shear deformable plate bending formulation discussed in Sec. 10.3. Some example problems are solved using MATLAB programs below. 4 Example 10.7.1 A simply supported square plate is subjected to a concentrated load at the center. Find the deflection of the plate using the shearSection 10.7 MATLAB Application 383 6 075 NORM. DEFLECTION 2 a 3 9 27 MESH IN A QUARTER PLATE Figure 10.6.2 Unifromly Loaded Circular Plate With Clamped Edge { 3 ELEMENTS 12 ELEMENTS 27 ELEMENTS Figure 10.6.3 Meshes for a Quarter of Circular Plate deformable displacement formulation. The size of the plate is 10 in. by 10 in. and its thickness is 0.1 in. It is made of a steel whose elastic modulus is 30 x 10° psi and Poisson’s ratio 0.3. The applied force is 40 Ib at the center. A quarter of the plate is modeled dne to symmetry and it is divided into 4 four-node elements (see Fig. 7.1.1). ‘The MATLAB program is written for the finite element analysis. Two point integration is used for the bending term while one point integration is used for the shear term for the selective integration technique. As far as the boundary conditions are concerned, two edges are simply supported and two edges are symmetric. As a result, nodes 1, 2 and 3 are constrained for 6, and w. Nodes 1, 4 and 7 are constrained for Oy and w. Nodes 3, 6 and 9 are constrained for 8, while nodes 7, 8 and 9 are constrained for Oy. The resultant constrained degrees of freedom are 1, 2, 3, 4, 6, 7, 9, 11, 12, 16, 20, 21, 23, 25, and 26, The external force is applied at node 9 with the third degree of freedom. Hence, the concentrated force is applied at the 27th degree of freedom of the load vector. Because of the quarter symmetry, a quarter of the force is applied to the load vector. The finite element solution gives the center deflection of 0.0168 in. while the analytical solution is 0.0169in.384 Plate Structures Chapter 10 SYMMETRY SIMPLE SUPPORT SYMMETRY SIMPLE SUPPORT Figure 10.7.1 A Quarter of Square Plate With 4 Elements %— % Example 10.7.1 % A simply supported square plate is subjected to a concentrated load % at the center. Find the deflection of the plate using 4 four-node % isoparametric elements of the shear deformable displacement. % formulation. The size of the plate is 10 in. by 10 in. and its % thickness is 0.1 in. It is made of steel and the applied % force is 40 Ib. % (see Fig. 10.7.1 for the finite element mesh) % % Variable descriptions % k = element matrix % kb = element matrix for bending stiffness % ks = element matrix for shear stiffness element vector system matrix yystem vector % disp = system nodal displacement vector % gcoord = coordinate values of each node % nodes = nodal connectivity of each element % index = a vector containing system dofs associated with each element % pointh = matrix containing sampling points for bending term % weightb = matrix containing weighting coefficients for bending term % points = matrix containing sampling points for shear term % weights = matrix containing weighting coefficients for shear term % bedof = a vector containing dofs associated with boundary conditions % beval = a vector containing boundary condition values associated with % the dofs in bcdof % kinmtpb = matrix for kinematic equation for bending % matmtpb = matrix for material property for bending % kinmtps = matrix for kinematic equation for shearSection 10.7 MATLAB Application % matmtps = matrix for material property for shear %— ——________. % ———— % input data for control parameters % number of elements % number of nodes per element % number of dofs per node % total number of nodes in system sdof=nnode*ndof; % total system dofs edof=nnel*ndof; % degrees of freedom per element emodule=30e6; % elastic modulus poisson=0.3; % Poisson’s ratio +=0.1; % plate thickness nglxb=2; nglyb=2; % 2x2 Gauss-Legendre quadrature for bending nglxs=1; nglys= % 1x1 Gauss-Legendre quadrature for shear % %—— % input data for nodal coordinate values % gcoord(i,j) where i-> node no. and j-> x ory ¢—$$$—$$$$__________ gcoord=[0.0 0.0; 2.5 0.0; 5.0 0.0; 0.0 2.5; 2.5 2.5; 5.0 2.5; 0.0 5.0; 2.5 5.0; 5.0 5.0}; % % % input data for nodal connectivity for each element % nodes(i,j) where i-> element no, and j-> connected nodes %— ee nodes=[1 25 4;2365;4587;569 8]; % Yo. ———$— ————“€~ % input data for boundary conditions eee bedof=[1 2 3 4679 11 12 16 20 21 23 25 26); % constrained dofs beval=zeros(1,15); % whose described values are zeros % es % initialization of matrices and vectors %— ff=zeros(sdof,1); % system force vector kk=zeros(sdof,sdof); % system matrix disp=zeros(sdof,1); % system displacement vector index=zeros(edof,1); % index vector kinmtpb=zeros(3,edof); % kinematic matrix for bending zeros(3,3); % constitutive matrix for bending sr08(2,edof); % kinematic matrix for shear 385386 Plate Structures Chapter 10 matmtps=zeros(2,2); % constitutive matrix for shear % %— % force vector %— f1(27) % ————————— — % computation of element matrices and vectors and their assembly % applied concentrated force % for bending stiffness % {pointb, weightb]=feglqd2(nglxb,nglyb); % % sampling points & weights matmtpb=fematiso(1,emodule,poisson)*t?/12; % % material property matrix % % for shear stiffness % [points, weights]=feglqd2(nglxs,nglys); % sampling points & weights .5*emodule/(1.0+poisson); % shear modulus % shear correction factor earm*shcof*t*[1 0; 0 1]; % material property matrix for iel=1:nel % loop for the total number of elements % for i=1:nnel nd(i)=nodes(iel,i); % extract nodes for (iel)-th element xcoord(i)=gcoord(nd(i),1); % extract x value of the nodes ycoord(i)=gcoord(nd(i),2); % extract y value of the nodes end % k=zeros(edof,edof); % initialization of element matrix kb=zeros(edof,edof); % initialization of bending matrix ks=zeros(edof,edof); % initialization of shear matrix % %— —— % numerical integration for bending term [ec for intx=I:nglxb x=pointb(intx,1); % sampling point in x-axis wtx=weightb(intx,1); % weight in x-axis for inty==]:nglyb y=pointb(inty,2); % sampling point in y-axis wty=weightb(inty,2) ; % weight in y-axis %Section 10.7 MATLAB Application [shape,dhdr,dhds}=feisoq4 (x,y); % compute shape functions and % % derivatives at sampling point % jacob2=fejacob2(nnel,dhdr,dhds,xcoord,ycoord); _-% Jacobian matrix % detjacob=det(jacob2); % determinant of Jacobian invjacob=inv(jacob2); % inverse of Jacobian matrix % [dhdx,dhdy]=federiv2(anel,dhdr,dhds,invjacob); _-% derivatives w.r.t. % % physical coordinate % kinmtpb=fekinepb(nnel,dhdx,dhdy); % bending kinematic matrix % %— % compute bending element matrix kb=kb+kinmtpb’*matmtpb*kinmtpb* wix*wty*detjacobs end end % end of integration loop for bending term % %- % numerical integration for bending term %— for intx=1:1 x=points(intx,1); % sampling point in x-axis wtx=weights(intx,1); % weight in x-axis for inty=1:nglys y=points(inty,2); % sampling point in y-axis wty=weights(inty,2) ; % weight in y-axis % [shape,dhdr,dhds]=feisoq4(x,y); % compute shape functions and % % derivatives at sampling point % jacob2=fejacob2(nnel,dhdr,dhds,xcoord,ycoord); % Jacobian matrix % detjacob=det(jacob2); % determinant of Jacobian inyjacob=inv(jacob2); % inverse of Jacobian matrix % [dhdx,dhdy]=federiv2(nnel,dhdr,dhds,invjacob); % derivatives w.r.t. % physical coordinate % kinmtps=fekineps(nnel,dhdx,dhdy,shape); % shear kinematic matrix % EE % compute shear element matrix ee ks=ks+kinmtps’*matmtps*kinmtps*wtx*wty*detjacob; 387388 Plate Structures Chapter 10 % end end % end of integration loop for shear term % ——— % compute element matrix ———— k=kb+! % index=feeldof(nd,nnel,ndof); % extract associated system dofs % kk=feasmbl1(kk,k,index); % assemble element matrices % end % —— % apply boundary conditions %—______—. [kk,ff]=feaplyc2(kk,ff,bcdof, beval); % % % solve the matrix equation displace=[num’ disp] % print nodal displacements % % : function [kinmtpb]=fekinepb(nnel,dhdx,dhdy) % % Purpose: % determine the kinematic matrix expression relating bending curvatures % to rotations and displacements for shear deformable plate bending % % Synopsis: % [kinmtpb]=fekinepb(nnel,dhdx,dhdy) % % Variable Description: % nmel - number of nodes per element % dhdx - derivatives of shape functions with respect to x % dhdy - derivatives of shape functions with respect to y % BcicceiceacecateuacectaicueHaie tatu euecaeueeneceennete % for i=I:nnelSection 10.7 MATLAB Application 389 is=i2+1; kinmtpb(1,i1)=dhdx(i); kinmtpb(2,i2)=dhdy(i); kinmtpb(3,i1)=dhdy (i); kinmtpb(3,i2)=dhdx(i); kinmtpb(3,i3)=0; end % % fanction [kinmtps]=fekineps(nnel,dhdx,dhdy,shape) % % Purpose: % determine the kinematic matrix expression relating shear strains % to rotations and displacements for shear deformable plate bending % % Synopsis: % [kinmtps]=fekineps(nnel,dhdx,dhdy,shape) % % Variable Description: % nnel - number of nodes per element % dhdx - derivatives of shape functions with respect to x % dhdy - derivatives of shape functions with respect to y % shape - shape function %— nel 1)*3+41; end % ——<—<—$ $$$ @& Example 10.7.2 The same square plate as that used in Example 10.7.1 is analyzed here. However, the boundary of the plate is clamped and the plate390 Plate Structures Chapter 10 is subjected to a uniform pressure of 2 psi. Using the same number of elements as before, determine the center deflection of the plate. Because of different boundary conditions and loads compared to the previous example, the following vectors substitute those in Example 10.7.1. Otherwise, the rest of the program is the same. The finite element result shows the center deflection of 0.0088 in. while the analytical solution is 0.0092 in. bedof=[1 23 4.56789 10 11 12 16 19 20 21 23 25 26); beval=zeros(1,19); 2 f1(21)=3.125; (24) =6.25; ff(27)=3.125;Problems 391 Figure P10.4 Problem 10.4 10.1 10.2 10.6 10.7 Problems Redo Example 10.7.1 for uniform pressure of 0.4 psi instead of the concentrated load using the computer programs. Compare the present solution to that in Example 10.7.1. Redo Example 10.7.2 for a center load of 200 Ib instead of the pressure load using the computer programs. Compare the present solution to that in Example 10.7.2. Redo Example 10.7.1 for a plate with two opposite edges simply supported and the other two opposite edges clamped. Redo Example 10.7.1 for a mesh shown in Fig. P10.4. Change the angle @ in the figure from 5 degrees to 30 degrees by an increment of 5 degrees. Compare the present solutions to that in Example 10.7.1. Solve a clamped circular plate which has elastic modulus of 200GPa, the radius of 0.2m, and thickness of 10mm. The plate is subjected to a center load of 2.0KN. Find the deflection using the computer programs for meshes shown in Fig. 10.6.3. Redo Prob. 10.5 for a simply supported plate. Find the deflection of a triangular shape of plate with simple support. The plate dimension is given in Fig. P10.7 and its thickness is 2mm. Its elastic modulus is 70GPA and it is subjected to a center force of 100N. Find the deflection of the plate using the computer programs. Redo Prob. 10.7 for the clamped plate.392 Plate Structures Chapter 10 015m Im Figure P10.7 Problem 10.7CHAPTER ELEVEN CONTROL OF FLEXIBLE STRUCTURES 11.1 Introduction The subject of flexible structures control consists of both dynamic analysis and control theory. Usually, these two disciplines are mingled together in such a way that we have to understand both disciplines to an equal extent in order to achieve our goal. The dynamic analysis of flexible structures is dominated by the finite element analysis as discussed in other chapters of this book. The control theory, on the other hand, is introduced in this chapter. The essence of each exemplary control theory is discussed in this chapter. In-depth discussion on control theories is available in a number of literature. The control theories are introduced here to help the readers of this book to understand the key features in conjunction with the finite element analysis of flexible structures. Prsented in Fig. 11.1.1 is a flow diagram representing the relationship between mathematical modeling and control system design for a given structural system. The mathematical modeling represented by finite element method has been discussed so far. The control system design in this chapter will mainly make use of the finite element modeling results. The control system design is demonstrated also using MATLAB. MATLAB m-files are generated in order to solve example problems. The example problems in this chapter do not need a specific MATLAB Toolbor. There are two distinct approaches for control system design. One is called the frequency domain approach or classical control technique and the other one is the time domain approach mainly adopted in the modern control technique. The frequency domain approach relies upon analytical tools, and is still popular in majority of existing control systems. For a given system, the frequency domain approach focuses on the relationship between input and output. The input to the system is modified to improve the output performance of a system. On the other hand, the modern control technique is motivated by the rapid advance of computational power. The system behavior is described by a set of variables over time domain. The control input then tries to control each variable in order to satisfy desired system responses. Each method has its own unique features. For the frequency domain approach, plenty of analytical results and tools are available. On the other hand, the modern control theory is easy to implement with the help of the abundant computational 393394 Control of Flexible Structures Chapter 11 StructuralSystem Mathematical Finite Element Heceing™ |L_Metnod Dynan Control age | Lh Desi Figure 11.1.1 Flow Diagram for Structural Analysis and Control software available. Modern control techniques impose another important issue on estimation of all degrees of freedom with a limited number of sensors. The number of sensors and actuators are usually less than the degrees of freedom of truncated (approximate) systems. Estimating all the flexible modes is so critical to designing an active control law in the time domain. Active control of flexible structures is mainly represented by vibration control using mechanical, electrical, and/or electromechanical devices. Inherent flexibility of the structure raises a number of issues in the area of active vibration control. The majority of flexible structures are distributed parameter systems. Therefore, they are essentially infinite dimensional dynamic systems. Obviously, the infinite dimensional systems are not practical for a control law design. Mathematical approaches like finte element analysis can be used to derive finite dimensional systems which closely duplicate the original infinite dimensional systems. Before we work on the dynamic analysis and control system design for flexible structures, we decide to introduce a basic stability theory. The stability theory is a key concept establishing the goal of a control law design. By stability, we mean the dynamic characeristic of a given dynamic system representing the behavior of dynamic motion of the system; for example, whether the motion is decaying or growing with respect to time. The Lyapunov stability theory has been considered as a background for under- standing the stability of a dynamic system. 11.2 Stability Theory ‘The Lyapunov stability theory is one of the most frequently referred tools for stability analysis and control system design of a dynamic system [37]. Definition of Stability Consider a general form of nonlinear system #=f(e,t), — 2(to) = 20 (11.2.1)Section 11.2 Stability Theory 395 20) 2) q 4 tee tt 4 4 t t (a) () Figure 11.2.1 Time History of Stable(a) and Asymptotically Stable(b) System where z is a vector and f(z,t) represents a general nonlinear function. The above system is stable in the Lyapunov sense with respect to the equilibrium state if for any given value ¢ > 0, there exists a number 8(¢,to) > 0 for which the ||z(t)]| < ¢ for all ¢ > to and |[z(to)]| < 6. The above condition implies that the magnitude of z(¢) remains within a finite small value in the presence of small initial perturbation. This definition includes also undamped pure oscillatory motions. Asymptotic Stability The system is asymptotically stable if it satisfies the stability condition and Timo [le(#){I = 0- The asymptotic stability implies that the state vector converges to the equi- librium point which is assumed zero at steady state in this case. The difference in the time response of both stable and asymptotically stable cases is displayed in Fig. 11.2.1. Lyapunov Second Stability Theory The second Lyapunov stability theory uses a nonnegative energy function for a given system. If the energy function decreases, then the system is stable toward an equilibrium point. In other words, the system energy is taken to be minimum at the equilibrium point. Theorem : Let V(z) be an energy function or a Lyapunov function. The system is stable if V(x) > 0 and V(2) < 0 for all values of 2. If the time rate of change of V(z) is less than zero, that is V(x) < 0, then the system is asymptotically stable. It is important to note that we cannot draw any conclusion about stability when a desired Lyapunov function is not found. In this case, we should try to find a Lyapunov function or apply Lyapunov’s instability theorem which is not discussed here. Another drawback of Lyapunov’s approach is that there is no systematic way of396 Control of Flexible Structures Chapter 11 finding an appropriate Lyapunov function candidate for stability proof. As it will be explained later in this chapter, the Lyapunov second stability theory is also used for a control law design; the control law tries to decrease a Lyapunov function suppressing the motion of a system. #& Example 11.2.1 Consider a set of coupled first order systems given by By = 22 — 24(2} + 23) fg = —2y — 2o(2? + 23) ‘We want to check stability of the system. The equilibrium points of the system can be obtained from a 2 =0 Based upon the equilibrium point, we select a trial Lyapunov function as V = 2? 423 which is always positive. The time derivative of the Lyapunov function in conjunction with the above set of equations turns out to be V = —2(x? + 23) Obviously, the V is guaranteed to be negative, and the system is asymptotically stable. t & Example 11.2.2 Consider a simple second order system whose governing equation is given as mg +cg+kg=0 where m > 0, c>0, & > 0. There are many ways to check the stability of the above system. In order to apply the Lyapunov theory, we transform the above equation into a set of first order equations as f= maiz = —cay — key where 21 = 4, 22 = ¢. Now we take a Lyapunov function as 1 V(e) = =(mz3 + kx?) 2 and the time derivative of the Lyapunov function becomes V(e) =maoe, + kee; aSection 11.2 Stability Theory 397 ‘Therefore, V(a) <0 as long as 2 # 0, and the system is asymptotically stable. Note that x2 can be instantaneously zero but converges to zero only at the steady equilibrium point. t Stability of a Linear First Order System Consider a linear first order system given as {a} = [Al{e} (11.2.2) where {z} is an n by 1 vector. In order to set up the stability condition of the system, we choose the following Lyapunov function U = {2} [Pl{z} (11.2.3) The time derivative of the Lyapunov function becomes U = {2} [P]{2} + {2} Pe} = {x}"(AT[P] + [PIA){2} (11.2.4) For stability we require ~(Q] = (Al? [P14 [PIA] (11.2.5) where {Q] is a positive definite matrix satisfying the property {z}7[Q]{z} > 0 for {2} #0 and [Q] = [Q]”. Thus, U —{2}"[Q{z} <0 (11.2.6) Therefore, we have the following theorem for stability of the linear system. Theorem: The linear system {2} = [A]{z} is stable if and only if there exists a positive definite [P] matrix which satisfies Eq. (11.2.5) for a given positive definite matrix [Q]. #& Example 11.2.3 Let us assume a two degree of freedom system @=tte=(% i]t In order to check the stability of the system, first we assume a [Q] matrix in Eq. (11.2.5) as a=[9 |398 Control of Flexible Structures Chapter 11 Once we have [Q], then we solve Eq. (11.2.5) [ode fe ILS Alf SIR 3] ‘The resultant [P] matrix, therefore, becomes 1f7 1 dear [i a] It is not difficult to check that [P] is positive definite, thus the system is stable. t Bounded Input Bounded Output(BIBO) Stability In the above definition of stability, we were concerned only about a system itself without including an external input, When there is a certain external input to the system, the stability of the system obviously should take the magnitude of the input into account. This is defined as, in general, the BIBO stability of the system as described below: When a system is under excitation by an external input with bounded magnitude, it is called BIBO stable if the output of the system is also bounded. Detailed mathematical description on BIBO stability will be provided later when we discuss the transfer function analysis of a system. 11.3 Stability of Multiple Degrees of Freedom Systems In the previous section, we discussed the stability theory, especially stability definition and Lyapunov function approach. Now, we want to discuss the stability of linearized multiple degrees of freedom system which is the main outcome of finite element analysis. Understanding the stability property of the multiple degrees of freedom system is so important before we make any attempt to design a feedback control law for a system. System without Damping Using the Lyapunov stability theory we want to analyze the stability of a linearized multiple degrees of freedom system. Let us consider an n dimensional finite dimensional dynamic system which is usually produced by finite element analysis. The governing equations of motion without damping are described by [M]{a} + [K]{a} = [F]{u} (11.3.1)Section 11.3 Stability of Multiple Degrees of Freedom Systems 399 where [M] is the system mass matrix, [K] is the stiffness matrix, {q} is the generalized coordinate vector, [F] is the input influence matrix, and {u} is the control input vector, In case there is no forcing function, the free vibrational motion satisfies {M]{4} + [K]{a} = 0 (11.3.2) and the solution to Eq. (11.3.2) is a pure sinusoidal motion as explained in Chapter 8 n {al} = Deegan} (113.3) Bea where {¢;} and w, are system parameters. The stability of Eq. (11.3.2) can be proved in various ways. One of them is the Lyapunov approach. Considering the fact that the total system energy(kinetic plus potential energy) is a direct indicator of system stability, a Lyapunov function candidate is suggested as U = Haya} + $a) Ika} 134) Note that the Lyapunov function is always positive (U > 0) since the mass and stiffness matrices satisfy {x}? [M]{x} > 0, {x}? [K]{x} > 0 (11.3.5) for {x} # 0, and they are symmetric ([M] = [MJ]? and [K] = [K]*). Next, the time derivative of the Lyapunov function in Eq. (11.3.4) in conjunction with Eq. (11.3.1) yields = % =tay"nta) + 1KHa)) ={a}" [Fi{u} (11.3.6) Without the external forcing input, that is {u} = 0, Eq. (11.3.6) becomes U= , or U=const (11.3.7) In other words, the energy is conserved, therefore the motion should be a pure sinusoidal type. It is important to note that the stability argument does not depend upon the system property itself. That is, the mass and stiffness matrices are dropped from the final expression of 7. This will be discussed again in the later part of this chapter when we deal with deriving a stabilizing control law for infinite dimensional systems. If we want to design the input {u} so that the system is stabilized, then one possible solution will be select {u} in such a way that u<0 (11.3.8)400 Control of Flexible Structures Chapter 11 Ul 1, Damper Figure 11.3.1 A Finite Element Beam Model with a Damper In other words, the energy decreases toward an equilibrium point with the judicious selection of the control input {u}. System with Damping A linearized multiple degrees of dynamic system with damping is described in the form (M]{a} + [D{4} + (K]{4} = [F{u} (11.3.9) where [D] is a nonnegative definite damping matrix, The above system is intuitively stable by the damping term introduced. In order to prove stability, we take a candidate Lyapunov function = 5a Ma) + (a) K1Ca)) (11.3.10) As it is shown, the Lyapunov function form is the same as that of Eq. (11.3.4). It is not surprising that both expressions are identical considering the Lyapunov function represents total energy (kinetic and potential energies) in both cases. The time derivative of the Lyapunov function becomes U = {a)" (Ma) + (Ka) (11.3.1) Using the governing equations of motion Bq. (11.3.9), we obtain U = {a}"(-(D{a} + [F{u}) (11.3.12) With the external control input ignored({u} = 0) u =-{4}"[Di{a} (11.3.1) Therefore, the time derivative of the Lyapunov function is a quadratic form in [D] and {q}. If the damping matrix [D} is positive definite with {4}7[D]{q} > 0, then the system is asymptotically stable. In case the damping matrix is only semidefinite, ie. {4}7[D]{4} > 0, then the asymptotic stability of the system is not guaranteed. In this case, we can use another technique in order to prove the stability.Section 11.4 Analysis of A Second Order System 401 # Example 11.3.1 Consider a finite element beam model in Fig. 11.3.1. ‘The material properties are given as EI = 1.112 x 104, p = 0.003, 1 = 20 with consistent units. A set of dampers is assumed at a nodal point in order to add damping to the system. The mathematical modeling of this system is given by [M] {4} + [D}{a} + [K]{a} = 0 where 0.0223 = 0-—«0.0039- 0.0093 t= | 2. 00571 0.0093 ~0.0214 =| 0.0039 0.0093 0.0111 —0.0157 0.0093 0.0214 0.0157 0.0286 0.2668 = 00.1334 0.6671 0000 aos 0 8.8946. 0.6671 2.2236 10 0 0.0 (k]=10° x | _9 334 —o.6671 0.1384 —0.6671]° PI=l0 0 ¢ 0 0.6671 2.2236 -0.6671 4.4473 0000 where c is the damping coefficient of the damper. When the Lyapunov function is taken as the total energy of the system, the time derivative of the Lyapunov function becomes U = ~{4}" Dita} ee. = -cg3 ‘As long as 43 # 0, U < 0 and the system is asymptotically stable. Even if the damping matrix has zero diagonal values, the system is still asymptotically stable since G3 # 0 except for the equilibrium point. t 11.4 Analysis of a Second Order System ‘A scalar second order system is frequently adopted as a reference explaining fundamental concepts of system responses. In fact, the majority of dynamical systems can be explained by using a scalar second order system. In this section, we want to introduce the natural frequency and damping ratio of a scalar system and key parameters associated with time responses. One of the most typical examples of the scalar second order system is a spring mass and damper system in Fig. 11.4.1. The governing equation of motion is given by m+ ck + ke = f(t) (11.4.1) where f(t) is the external force applied to the mass. Dividing both sides by the mass yields B+ (c/m)z + (k/m)z = f(t)/m (11.4.2)402 Control of Flexible Structures Chapter 11 ZZ ST, Figure 11.4.1 An Example of a Second Order System At this point, we define two parameters wn = Vk/m, 6 = 6/eer (11.4.3) where wp, is the natural frequency and ¢ is the damping ratio. In addition, c., = 2Vmk is defined as the critical damping ratio. As a consequence, Eq. (11.4.2) can be rewritten as E+ wad + wre = w?2 F(t) (11.4.4) where we used f(t)/é = F(t). In order to derive the solution, we use the Laplace transform technique ignoring initial conditions temporarily. This is also motivated by the fact that, for linear systems, the stability condition is independent of the initial conditions. (5? + 2€wns +w2)X(s) = w2F(s) (11.4.5) ‘Therefore, X(8)/F(s) = wh /(s? + 2uns + wn) (11.4.6) where X(s) = L[z(t)] and F(s) = L[F(t)] are Laplace transforms of z(t) and F(t), respectively. The equation obtained by setting the denominator of the transfer function equal to zero is called characteristic equation. The characteristic equation of the above transfer function, therefore, is given by 3? + uns +w2 =0 (11.4.7) and the solution is given by 81,2 = Gn tun C2 —1 (11.4.8) The above characteristic root is plotted on a complex plane in Fig. 114.2. The dynamic behavior is dependent upon the magnitude of the damping ratio, ¢ and the natural frequency wn.Section 11.4 Analysis of A Second Order System 403 Figure 11.4.2 Characteristic Roots of the Second Order System ‘As a special case, we analyze a step response of the system by selecting F()=1, t>0 or F(s)=1/s (11.4.9) that is, @ unit step input, By substituting F(s) into the above equation, the output can be found by the inverse Laplace transform technique 2(t) = L~![X(s)]. The result turns out to be dependent upon the magnitude of the damping ratio ¢. i). Underdamped case, 0 < ¢ < 1 In this case, s1,2 are complex conjugate, and the motion turns out to be a damped oscillatory one. entunt z(t) =1- in(wat + 4) (11.4.10) where ¢ = tan-*/(1—(4)/¢ and wa = V(1—C?)w, is the damped natural frequency. ii). Critically damped case, ¢ = 1 In this case, 61,2 ate repeated real numbers. The motion is monotonically increasing torward the steady state value x(t) =1—e7-%""(1 + wat) (11.4.1) iii). Overdamped case, ¢ > 1 In this case #1, are all negative real. The motion is also monotonically increasing toward the steady state —Cwat (62-1 wat 6/0? =1)wat ae ( . © ) (114.12) a1 AAs Vern C—O alin Three different motions are presented in Fig. 11.4.3, As seen in the figure, the usual response trends are exponentially decaying motions dictated by e~*s# superimposed by damped sinusoidal motion with damped frequency, w,. In particular, the constant 7 = 1/Cwn is called time constant which represents how long it takes for404 Control of Flexible Structures Chapter 11 aft) Under damped, ¢ <0 Critically damped, ¢ = 1 time Figure 11.4.3 Step Responses with Different Damping Ratios the response to reach a certain level from an initial condition. The shorter the time constant the quicker the response tends to be. For an underdamped motion, some parameters are introduced characterizing the transient response of the motion. Those parameters are sometimes used to prescribe design specifications. In Fig. 11.4.4, different labels are used to denote those specifications. i) Rise time, ty: The rise time is the required time for the response to start from zero value and cross the unit steady state value. The rise time is found from (1.4.13) It turns out that (114.14) ii) Peak time, tp: Peak time is the instance when the response reaches a maximum value. The peak time can be obtained as (11.4.15) which produces (114.16)Section 11.4 Analysis of A Second Order System 405 x(t) Figure 11.4.4 Key Parameters in the Step Response iii) Mazimum overshoot, My: The maximum overshoot represents the amount of maximum deviation of the response from the steady value. From Fig. 11.4.4, the maximum overshoot satisfies My = 2(tp) - 1 =e CIVIC (1.4.17) iv) Settling time, t,: The step response of the underdamped system experiences a transient response finally reaching a steady state value. The settling time represents the amount of time it takes for the response to stay within a certain band prescribed around the steady state value. The size of the band can be selected as, for example, 2% or 5%. For 2% band, the settling time turns to be approximated by the multiple of the time constant, Tr. 4 tsar oe (1.4.18) @ Example 11.4.1 A second order system is given by X()/F() = ay Ris ue where w=2.0 (rad/sec) and (=0.2. The unit step response results are calculated by calling fesecnd(¢, wn) command. [tp,teste, Mp] = (1.603, 1.772, 10, 0.527] function [t-p, t-z, ts, M-p]=fesecnd(zeta, w-n)406 Control of Flexible Structures Chapter 11 Re Purpose: % The function subroutine fesecnd.m calculates dynamic characteristics % — ofa typical standard second order system. % won? % H(s)= — % 8°242*zetatwon*s+wn"2 % % Synopsis: % [t-p, tx, M-p, t.s]=fsecond(zeta, wn) % % Variable Description: % Input parameters : zeta - damping ratio % wn - natural frequency % Output parameters : t-p- peak time, tx rise time % t.s- settling time, M-p- maximum overshoot, ee —————————————— w_d=sqrt(1-zeta’2)*w-n; % Calculate undamped natural frequency t_p=pi/w-d; % Calculate peak time t_r=atan2(sqrt(1-zeta”2), -zeta); % Calculate rise time ts=4/zeta/wn; % Calculate settling time M-p=exp(-zeta*pi/sqrt(I-zeta"2)); -% Calculate maximum overshoot ——————— ‘The same analysis and definition can be applied to multiple degree of freedom systems which may be generated by finite element analysis. From the original governing equation (M]{4} + [D]{4} + [}{a} = [F]{u} (1.4.19) and assuming Rayleigh damping with [C] = a[M]+£[K], we obtain modal coordinate form governing equation f+ wih turn =f, 1=1,2,. (1.4.20) Now for each modal coordinate parameters(w;,¢;) we can check the dynamic charac- eristics. 11.5 State Space Form Description In general, the equations of motion of dynamic systems are described by second order differential equations. Finite element modeling of dynamic systems also results in second order differential equations of motion. The second order differential equations cover a generic class of dynamic systems. The analytical solution of secondSection 11.5 State Space Form Description 407 u oD ay oo Figure 11.5.1 Graphical Representation of the State Space Equation order equations of motion is essentially equivalent to solving ordinary differential equations. On the other hand, the first order state space form description of dynamic systems has certain advantages over the second order form description. The second order equations can be transformed into first order equations and the first order forms also can be transformed into second order equations. Majority of existing computer software tools are written for the first order systems. This is due to the inherent nature of first order equations which are more convenient for numerical computations. Another significant advantage of the first order form descriptions is that we can analyze the equations in an explicit form. Consider a linearized second order dynamic system (M}a} + (D]}{4} + [k}a} = [FHu} (11.5.1) In order to write the above equation in the first order form we introduce a vector which is usually called the state vector fe} = {3} (115.2) Now we have the following relationship ya ffale 4 = 4 {3}= {toa - neta oer sten} = 0 I q 0 =. are eax] {3} : [ian] oe = (Alte) + 18a} In other words, we rewrite the original second order differential equation in the first order form by introducing the state vector {z}. The state vector and associated properties constitute the so-called state space. The state space and related subjects are well described in Refs. [38-41]. ‘The size of the first order system, however, increases by twofold compared to the original second order system. This may seem to be a drawback; however, the modern computational capability resolves this concern to a considerable extent. As mentioned earlier, the first order form has certain advantages being adopted in the majority of engineering applications. Graphical representation of the first order state space equation is presented in Fig. 11.5.1.408 Control of Flexible Structures Chapter 11 Eigenvalue Problem and Free Response Let us consider a case without an external input, that is {u} = 0, sometimes called the autonomous system. Thus, Eq. (11.5.3) reduces to {e =[Al{z} (11.5.4) In order to find the solution {2(t)} to the above equation, we assume {x(t)} = ce*#{$} (11.5.5) where c is an arbitrary constant and {¢} is a vector of consistent size with {x(t)}. ‘Then we substitute Eq. (11.5.5) into Eq. (11.5.4) arriving at e(AL — [A}e**{4} = 0 (11.5.6) In other words, (I-A) {6} = 0 (11.5.7) Since {4} should be a nonzero vector, for the existence of a nontrivial solution, it should follow Pr- [Al] =0 (115.8) The above equation can be solved for 2 for the given system matrix [A]. There are n ’s as the size of [A]. For an arbitrary i** );, we can rewrite Eq. (11.5.7) as Ai{di} = [A] {oi} (11.5.9) Equation (11.5.9) is called the eigenvalue problem, which is a crucial concept in linear dynamic system analysis. Even if we eliminate the control input temporarily for the eigenvalue analysis, the eigenvalues and eigenvectors are used for stability analysis and computing mode shapes of the system. For a given system, there may exist repeated eigenvalues, for example, a system with rigid body motion such as pure translational and rotational motions. This situation is treated by somewhat different approaches. Momentarily, the analysis is restricted to the case where the eigenvalues are all distinct. Let us write the eigenvalue problem for each index Ar{d1} = [AH Gi} Aa{da} = [AN $2} : (11.5.10) Ant{dn} = [A]{dn} The above set of equations can be combined into a single matrix equation form (PU) = (WItA) (115.11)Section 11.5 State Space Form Description 409 where [U/] is a matrix whose columns consist of eigenvectors and [I] is a diagonal matrix for which each diagonal term consists of eigenvalues. [0] = [{91}, {2},---.{4n}], [A] = diag [Ai], Providing that the eigenvalues are all distinct, Bq. (1.5.11) turns into a relationship [38-41] an | (11612) [4] = (UP TAY] (1.5.13) That is, the matrix [A] can be rewritten as a combination of a matrix which consists of eigenvectors and a diagonal matrix of eigenvalues. As discussed in the above, when there are repeated eigenvalues, we should use a modified form of equation, Equation (11.5.13) is named as similarity transformation of A. Note that the eigenvalue problem is invariant under the similarity transformation JAZ = (U}-AN] = |[0-* (7) - (TAN) = [luz ~ fal |(01] = |Ar-[A]] (11.5.14) where we used |[U]| = 1/|[U]-?]. ‘As we might remember, the eigenvaule problem for second order systems in Chapter 8 can be similarly applied to a first order system in this case. The solution of the eigenvalue problem leads us to the analytical expression for {2(t)} {a(t)} = cre "{41} + e2e*"{ ba} +--+ ene{bn} (11.5.15) The constants (c1,c2,---,¢n) are obtained from the initial condition. Equation (1.5.15) tells us that once we compute eigenvalues and eigenvectors, then we obtain the expression for the response with respect to the initial condition. & Example 11.5.1 Let us consider a finite element model for a beam with only one element as shown in Fig. 11.5.2. The mass and stiffness matrices for this system are ee a [M1] = 495 —22h 4h? nB |-6h 4h? rh 156 —22h EI[ 12 -6h ] = | where p = 0.002kg/m is the linear mass density, EI = 10 Nm? is the beam rigidity, and h = 1 m the element length. The second order system is converted into the first order system in accordance with Eq. (11.5.3). The result is 0 0 1.0000 0 l= 0 0 0 1.0000 = 1630x104 —4.8 x 104 0 0 5.04 x 105 —3.69 x 10° 0 0410 Control of Flexible Structures Chapter 11 Now we have a MATLAB built-in function /V,DJ=eig([A]) command, as explained in Chapter 1, in order to compute the eigenvalue and eigenvector of [A]. Consequently, Dia = £550.35%, gq = 15.86% 2.3635 x 1074 2.3635 x 1074 _ J 1.8016 x 10-% _ J 1.8016 x 10-9 {d= ) “orzo f? {2}= 4) "_orsors 0.9915i —0.9915i —1.0516 x 10-? 1.0516 x 10-7 — J -1.4485 x 10-? _ J -1.4485 x 107? {ta}= —0.5874i 2 ae 0.58741 —0.8091i 0.8091¢ The time response of the system due to the initial condition can be written as {x(t)} = cr{dr}e™! + co{da}e™* + cs{ds}e™*t + calba}e In order to find the constants(c;,C2,¢3,¢4), We assume an initial condition as 0.05 —0.01 0.1 0.0 {2} = ‘Therefore, the constants are calculated as C1, ¢2 = —26.719 + 8.4788 x 10771, 3, e4 = —2.9779 + 0.10391 Substituting the constants into {z(t)} and using the famous Euler’s formula as (38) e'? = cos(0) + isin(0) we obtain the final form for {2(t)} analytically —1.2630 x 10-? —4.0079 x 10-5 9.6274 x 107? —3.0551 x 10-* | {2(} =) 75 062 x 10-2 7 °28(550:35t) + aes sin(550.35t)-+ ~1.6813 x 107? 5.2984 x 10? 6.2673 x 107? 2.1852 x 107? 8.6328 x 10-? 3.0100 x 10-3 |. 1.2206 x 10-1 f °08(55-86t) + 4°" 5 syog fin (55.86¢) 1.6813 x 107! —4.8221Section 11.5 State Space Form Description 4u1 Figure 11.5.2 A Finite Element Model with a Single Element Forced Response and State Transition Matrix The solution to the first order equation Eq. (11.5.3) can be also found by using the Laplace transform technique X(s) — {2(0)} = [A]X(s) + [B]U(s) (1.5.16) where X(s) = L{{z(1)}] is the Laplace transform of {2(#)}, and U(s) = L[{u(t)}] is the Laplace transform of {u(t)}. Alternatively, X (8) = (sI ~ [A])~'2(0) + (87 — [A])“'[B]U(s) (1.5.17) In order to find the time response, we take the inverse Laplace transform of X(s). First, the inverse Laplace transform of (sI — [4])~! should be evaluated. If a is a scalar, we know L((s —a)7!] = e* (11.5.18) Generalization of Eq. (1.5.18) to a matrix [A], we can derive a similar relationship as shown below. I-((sI — [A])~4] = el4lt (11.5.19) Therefore, {x(t)} = 4lt{2(0)} + f[ fAlC—)[B]{u(r)}dr (11.5.20) 0 where the second term on the right-hand represents a convolution integral. The response consists of two parts; one is due to the initial condition ({2(0)}) and the other one is due to the control input ({u}). It is not as straightforward to understand ell as the scalar case e%. In order to analyze ell, first. we introduce the Talyor series expansion. (aye? 2! [ape ef4lt = [7] + [Ale t+ 3i at (11.5.21) ‘The first fundamental question is whether the infinite series converge or not. The answer is yes from a physical intuition; the factorial term in the denominator412 Control of Flexible Structures Chapter 11 dominates the exponential term in the numerator. Efficient numerical algorithms have been developed and a MATLAB built-in routine expm is also available. Another interesting property can be derived from Eq. (11.513) which calculates the exponential of a matrix efit = [uy telI[U] (1.5.22) where eA = diagle™], §=1,2,...,0 The proof follows as ofAlt = [1] + [A]t + [A]?t?/2! + [A]?e2/31+ + = UP) + (OP Ae) + (UP AOU Tao)? /2 + (UI TAO CO AMO) (OT TATU) F/3! + +» = (U)*(U1 + [Ale + (AP 7/2! + [ASH9/31+ + [0] = [uy tel] (1.5.23) Therefore, the eigenvalue solution can be used again to calculate el4lt, Let’s go back to the first order equation without the control input term. Thus, {2} = [A]{z} (11.5.4) In order to derive the solution, we assume {2(t)} = [®(t, r){2(7)} (1.5.24) where [6(t, 7)] is the so-called state transition matrix which relates the state variable at different instants, that is, {2(t)} with {2(r)} for ¢ > 7. Obviously, [6(¢,7)] is a time varying matrix. Now substitute Eq. (11.5.24) into Eq. (11.5.4). (BE, T)He(7)} = (ANE, Ne(7)} (1.5.25) Therefore, | [®(,7)] = [AI@E, 7)] (1.5.26) By solving Eq. (11.5.26) we can derive the analytical expression of {x(t)}. The above differential equation is combined with the initial condition of [®(¢,7)] by noting that {2(r)} = O(r,7){2(7)} (1.5.27) In other words, [O77 =7 (11.5.28) and the combination of Eqs. (1.5.26) and (1.5.28) constitutes a matrix differential equation. Obviously, a numerical integration technique can be applied to the differential equation. There are other useful properties of [(t, r)] [38-41]. @ [8G,7)] = (OE, 42)G@2,7)], t

You might also like

• Synthesis Lectures On Mechanical Engineering Shuvra Das An Introduction

  

  No ratings yet

  Synthesis Lectures On Mechanical Engineering Shuvra Das An Introduction

  228 pages
• Advanced Mathematics &mechanics Applications Using MATLAB

  

  No ratings yet

  Advanced Mathematics &mechanics Applications Using MATLAB

  10 pages
• Introduction To The Mathematics of The Finite Element Method (Kathryn Gillow)

  

  No ratings yet

  Introduction To The Mathematics of The Finite Element Method (Kathryn Gillow)

  58 pages
• Finite Element Analysis Simply Explained: T. C. Kennedy Oregon State University July 2016

  

  No ratings yet

  Finite Element Analysis Simply Explained: T. C. Kennedy Oregon State University July 2016

  23 pages
• Finite Element Analysis

  

  No ratings yet

  Finite Element Analysis

  25 pages
• Coastal Engineering Sediment Models

  

  No ratings yet

  Coastal Engineering Sediment Models

  19 pages
• Curve Fitting in Matlab

  

  No ratings yet

  Curve Fitting in Matlab

  216 pages
• Rlabz Tech: Training - Workshops - Seminar - Consultancy

  

  No ratings yet

  Rlabz Tech: Training - Workshops - Seminar - Consultancy

  7 pages
• 1989 Book ViscousFlowApplications

  

  No ratings yet

  1989 Book ViscousFlowApplications

  195 pages
• The Finite Element Method-An Introduction

  

  No ratings yet

  The Finite Element Method-An Introduction

  125 pages
• Introduction To FEM

  

  No ratings yet

  Introduction To FEM

  439 pages
• Structural Material Models Manual

  

  No ratings yet

  Structural Material Models Manual

  121 pages
• Toro 4-ADER

  

  No ratings yet

  Toro 4-ADER

  63 pages
• Lecture 09 - Sequential Quadratic Programming

  

  No ratings yet

  Lecture 09 - Sequential Quadratic Programming

  4 pages
• FEM Lecture Notes by Peter Hunter, Andrew Pullian

  

  No ratings yet

  FEM Lecture Notes by Peter Hunter, Andrew Pullian

  153 pages
• (Chen, Changseng Et Al, 2006) An Unstructured Grid, Finite Volume Coastal Ocean Model (FVCOM) PDF

  

  No ratings yet

  (Chen, Changseng Et Al, 2006) An Unstructured Grid, Finite Volume Coastal Ocean Model (FVCOM) PDF

  324 pages
• Computational Methods Handbook

  

  No ratings yet

  Computational Methods Handbook

  46 pages
• 103106118

  

  No ratings yet

  103106118

  343 pages
• Matlab Manual 2020 21 Mechanical

  

  No ratings yet

  Matlab Manual 2020 21 Mechanical

  58 pages
• MATLAB 2023b Partial Differential Equation Toolbox

  

  No ratings yet

  MATLAB 2023b Partial Differential Equation Toolbox

  2,258 pages
• Changing Plot Style and Color: Erin Case

  

  No ratings yet

  Changing Plot Style and Color: Erin Case

  54 pages
• Finite Element Method (FEM) - Comsol

  

  No ratings yet

  Finite Element Method (FEM) - Comsol

  33 pages
• Finite Element Methods Book List

  

  No ratings yet

  Finite Element Methods Book List

  7 pages
• OpenSees Command Collection in Compact

  

  100% (2)

  OpenSees Command Collection in Compact

  36 pages
• Data Visualization With Seaborn

  

  No ratings yet

  Data Visualization With Seaborn

  3 pages
• MATLAB Mathematics (MathWorks)

  

  No ratings yet

  MATLAB Mathematics (MathWorks)

  768 pages
• Simulink Tutorial PDF

  

  No ratings yet

  Simulink Tutorial PDF

  9 pages
• Wavelet Transforms and Their Recent Applications in Biology and Geoscience

  

  No ratings yet

  Wavelet Transforms and Their Recent Applications in Biology and Geoscience

  310 pages
• LS Dyna Manual

  

  No ratings yet

  LS Dyna Manual

  2,538 pages
• Finite Element Method

  

  No ratings yet

  Finite Element Method

  172 pages
• HLLC Riemann Solver: Eleuterio F. Toro Laboratory of Applied Mathematics University of Trento, Italy Toro@ing - Unitn.it

  

  No ratings yet

  HLLC Riemann Solver: Eleuterio F. Toro Laboratory of Applied Mathematics University of Trento, Italy Toro@ing - Unitn.it

  50 pages
• Geochemical Modelling of Groundwater, Vadose and Geothermal Systems PDF

  

  No ratings yet

  Geochemical Modelling of Groundwater, Vadose and Geothermal Systems PDF

  76 pages
• Introduction To Finite Element Method

  

  No ratings yet

  Introduction To Finite Element Method

  16 pages
• PyFEM 1.0 User Manual for Installation and Usage

  

  No ratings yet

  PyFEM 1.0 User Manual for Installation and Usage

  22 pages
• Seaborn for Data Visualization

  

  No ratings yet

  Seaborn for Data Visualization

  22 pages
• Matlab Finite Element Modeling For Materials Engineers Using MATLAB

  

  No ratings yet

  Matlab Finite Element Modeling For Materials Engineers Using MATLAB

  74 pages
• The Finite Element Method Using MATLAB - Kwon and Bang

  

  60% (5)

  The Finite Element Method Using MATLAB - Kwon and Bang

  527 pages
• Finite Element Method Using Matlab PDF

  

  No ratings yet

  Finite Element Method Using Matlab PDF

  527 pages
• Finite Element Method Using Matlab

  

  100% (1)

  Finite Element Method Using Matlab

  527 pages
• The Finite Element Method Using MATLAB Young W. Kwon, Hyochoong Bang PDF

  

  100% (1)

  The Finite Element Method Using MATLAB Young W. Kwon, Hyochoong Bang PDF

  590 pages
• MATLAB and C Programming For Trefftz Finite Element Methods 1st Edition Qing-Hua Qin Digital Download

  

  100% (2)

  MATLAB and C Programming For Trefftz Finite Element Methods 1st Edition Qing-Hua Qin Digital Download

  147 pages
• Numerical and Analytical Methods With MATLAB 1st Edition William Bober Newest Edition 2025

  

  No ratings yet

  Numerical and Analytical Methods With MATLAB 1st Edition William Bober Newest Edition 2025

  104 pages
• Gautam Dasgupta (Auth.) - Finite Element Concepts - A Closed-Form Algebraic Development-Springer-Verlag New York (2018)

  

  100% (2)

  Gautam Dasgupta (Auth.) - Finite Element Concepts - A Closed-Form Algebraic Development-Springer-Verlag New York (2018)

  358 pages
• Full Download Finite Element Simulations Using ANSYS Second Edition. Edition Alawadhi PDF

  

  100% (5)

  Full Download Finite Element Simulations Using ANSYS Second Edition. Edition Alawadhi PDF

  84 pages
• 4979dynamics in Engineering Practice Eleventh Edition Childs Full Chapters Instanly

  

  No ratings yet

  4979dynamics in Engineering Practice Eleventh Edition Childs Full Chapters Instanly

  170 pages
• Introduction To Finite Element Analysis For Engineers 2nd Edition Saad A. Ragab & Hassan E. Fayed Digital Version 2025

  

  100% (1)

  Introduction To Finite Element Analysis For Engineers 2nd Edition Saad A. Ragab & Hassan E. Fayed Digital Version 2025

  111 pages
• Finite Element Analysis With Numeric and Symbolic Matlab (J. E. Akin) (Z-Library)

  

  No ratings yet

  Finite Element Analysis With Numeric and Symbolic Matlab (J. E. Akin) (Z-Library)

  662 pages
• Calfem 3.3 PDF

  

  No ratings yet

  Calfem 3.3 PDF

  280 pages
• Pavel Sumets - Computational Framework For The Finite Element Method in MATLAB® and Python-CRC Press (2022)

  

  100% (3)

  Pavel Sumets - Computational Framework For The Finite Element Method in MATLAB® and Python-CRC Press (2022)

  182 pages
• Advanced Mathematics and Mechanics Applications Using MATLAB 3rd Ed Edition David Halpern Instant Download

  

  No ratings yet

  Advanced Mathematics and Mechanics Applications Using MATLAB 3rd Ed Edition David Halpern Instant Download

  163 pages
• Extended Finite Element and Meshfree Methods 1st Edition Timon Rabczuk PDF Download

  

  100% (8)

  Extended Finite Element and Meshfree Methods 1st Edition Timon Rabczuk PDF Download

  81 pages
• Advanced Finite Element Simulation With MSC Marc Application of User Subroutines 1st Edition Zia Javanbakht Full Digital Chapters

  

  No ratings yet

  Advanced Finite Element Simulation With MSC Marc Application of User Subroutines 1st Edition Zia Javanbakht Full Digital Chapters

  118 pages
• Calfem Manual

  

  No ratings yet

  Calfem Manual

  285 pages
• Calfem

  

  No ratings yet

  Calfem

  285 pages
• Finite Element Computations in Mechanics With R: A Problem-Centered Programming Approach 1st Edition Khameel Bayo Mustapha Full Digital Chapters

  

  No ratings yet

  Finite Element Computations in Mechanics With R: A Problem-Centered Programming Approach 1st Edition Khameel Bayo Mustapha Full Digital Chapters

  93 pages
• Matlab Fem

  

  No ratings yet

  Matlab Fem

  45 pages
• Matlab Finite Element Coding Guide

  

  100% (3)

  Matlab Finite Element Coding Guide

  45 pages
• Matlab for Finite Element Coding

  

  No ratings yet

  Matlab for Finite Element Coding

  45 pages
• MATLAB Finite Elements Guide

  

  0% (1)

  MATLAB Finite Elements Guide

  1 page