Mesh Deformation Using
Radial Basis Functions
Neil Matula
Department of Aerospace Engineering
Texas A&M University
1876
November 13, 2012
Introduction
Theory
Results
Conclusion
Outline
Introduction
Problem Statement
Spring Analogy
Comparison
Theory
Definition
Interpolation
Constraints
Matrix Formulation
Evaluation
Results
RBF vs. Spring
Conclusion
Neil Matula
Mesh Deformation Using Radial Basis Functions
2 / 13
Introduction
Theory
Results
Conclusion
Problem Statement
In aeroelastic simulations, geometry changes after
application of structural solver.
Two options:
Remesh - Guarantees a high quality mesh, but expensive.
Deform Mesh - Cheap, but does not guarantee a quality
mesh.
Neil Matula
Mesh Deformation Using Radial Basis Functions
3 / 13
Introduction
Theory
Results
Conclusion
The Spring Analogy
Neil Matula
Edges and angles replaced with fictitious linear and
torsional springs
Move boundary, get equilibrium
Guarantees a certain amount of quality
Trouble with certain meshes and large deflections
Mesh Deformation Using Radial Basis Functions
4 / 13
Introduction
Theory
Results
Conclusion
Comparison
Connectivity
Mesh Type
Computation
Quality Enforced
Neil Matula
Spring
Used
Unstructured
Expensive
Yes
TFI
Used
Structured
Cheap
No
Mesh Deformation Using Radial Basis Functions
RBF
Not Used
Either
Medium
No
5 / 13
Introduction
Theory
Results
Conclusion
Definition
A radial basis function (RBF) is a function of distance from the
origin. They are frequently used to build function
approximations (hence, basis function). Some commonly used
RBFs with global support are the following:
Name
Thin Plate Spline
Multiquadratic Biharmonics
Inverse Multiquadratic Biharmonics
Quadratic Biharmonics
Inverse Quadratic Biharmonics
Gaussian
Neil Matula
Mesh Deformation Using Radial Basis Functions
Function
x2 log(x)
a2 + x2
1
a2 +x2
2
1+x
1
1+x2
x2
6 / 13
Introduction
Theory
Results
Conclusion
Interpolation
We seek to interpolate the displacements of discrete boundary
nodes to the interior of the domain using the following:
s(x) =
N
X
i (ri ) + h(x)
i=1
h(x) = 1 + 2 x + 3 y + 4 z
ri = ||x xi ||
x = (x y z)T
x
xi
ri
s
i
N
h
Neil Matula
Position
Position of boundary node i
Distance between the current position and boundary node i
Radial basis function
The interpolated displacement (scalar)
Linear polynomial coefficients
Number of boundary nodes
Linear polynomial
Mesh Deformation Using Radial Basis Functions
7 / 13
Introduction
Theory
Results
Conclusion
Constraints
We have the following two constraints to determine the
necessary coefficients:
Neil Matula
s(xi ) = d(xi ), i = 1..N
This indicates that we want to recover exactly the
displacements on the boundary.
PN
T
i=1 i (1, xi , yi , zi ) = 0
This states that all polynomials of degree less than or equal
to h are interpolated exactly.
Mesh Deformation Using Radial Basis Functions
8 / 13
Introduction
Theory
Results
Conclusion
Matrix Formulation
We can assemble these constraints into a linear system in the
following manner:
PT
Mij = kxi xj k
Neil Matula
!
=
1
1
P = .
..
x1
x2
..
.
y1
y2
..
.
z1
z2
..
.
1 xN
yN
zN
Mesh Deformation Using Radial Basis Functions
9 / 13
Introduction
Theory
Results
Conclusion
Evaluation
This [N + 4] [N + 4] matrix equation can be solved for
coefficient vectors and . Then all mesh points on interior
can be evaluated using
s(x) =
N
X
i (ri ) + 1 + 2 x + 3 y + 4 z
i=1
Since s(x) is scalar-valued function, each of three displacement
directions must be treated separately.
Neil Matula
Mesh Deformation Using Radial Basis Functions
10 / 13
Introduction
Theory
Results
Conclusion
RBF vs. Spring
Radial Basis Functions
Neil Matula
Spring Analogy
Mesh Deformation Using Radial Basis Functions
11 / 13
Introduction
Theory
Results
Conclusion
Conclusion
Conclusion
The spring algorithm produces superior mesh quality in
general.
Under certain circumstances, such as large deflections and
poor starting meshes, RBF will perform better.
RBF is also cheap and applicable to various mesh types.
Neil Matula
Mesh Deformation Using Radial Basis Functions
12 / 13
Introduction
Theory
Results
Conclusion
Questions?
Neil Matula
Mesh Deformation Using Radial Basis Functions
13 / 13
