QUATTRO RDOF Use: Software Guidelines
USE OF ROTATIONAL DEGREES OF FREEDOM IN STRUCTURAL
DYNAMICS ANALYSIS: SOFTWARE GUIDELINES
PUBLIC
Contract nr.: BRPR-CT-0540
Project nr.: BE97-4184 QUATTRO
Title: Quantitative treatment and testing of rotational
degrees of freedom
Project co-ordinator: Instituto Superior Tecnico
Partners: University of Trieste
Chalmers University
Imperial College
Bosch
LMS International
Starting date: 01-01-1998 Duration: 33 months
PROJECT FUNDED BY THE EUROPEAN
COMMUNITY UNDER THE BRITE/EURAM
PROGRAMME
QUATTRO RDOF Use: Software Guidelines
1. Introduction
Since the final aim of the QUATTRO project is to use the project findings in the industrial
engineering practice, an important issue is to study the consequences of the project with
respect to the currently used software tools for Computer Aided Testing (CAT) and
Computer Aided Engineering (CAE).
To do this, a study was made of the various CAT/CAE applications where RDOF
information was found to be relevant, and the software requirements were derived in order
to allow the developed sensors and analysis methods to be used consistently for these
applications. The relevant aspects are:
- The support of the used data entities,
- The support in modal test-data analysis tools
- The support in transfer path/substructuring analysis tools
- The support in numerical analysis tools
Furthermore, next to the pure RDOF issue, the effect of proper residual compensation was
detected and solutions proposed..
Below, an overview of the conclusions is provided.
2. Data entities
Most software packages for CAT/CAE describe the processed data by means data types
and attribute a unit to these data. While in Finite Element Analysis, rotational degrees of
freedom are generally supported, with all related support in terms of units and data
headers, this is less the case in experimental analysis. So, next to the linear quantities
(and units), also their rotational equivalent must be supported in the data structures.
This not only requires the correct data descriptors and units for the measurement
transducers, calibration values and signal data blocks, but also all depending operators
(like the calculation of the correlation function, the frequency response function etc.) must
consistently follow the nature of the basic data types.
For example, for the FRF, the 4 cases T/T, R/T, T/R and R/R must automatically result
when making the FRF calculations. This requires the proper bookkeeping of dimensions
and each data operator has to be compatible with this bookkeeping.
Furthermore, in experimental analysis, not only the basic rotational quantities, but also the
first and second order derivatives are used (angular velocity, angular acceleration). For
example, a recalculation to the zero-order base quantity is normally performed before a
modal analysis is executed.
An overview of the various units as should be supported in modal analysis (including vibro-
acoustic analysis) is given in Table 1.
QUATTRO RDOF Use: Software Guidelines
Possible input Internal conversion Case
x x s2 1. T
f f kg T
x&
f
&&
x
f
θ θ rads 2 2. R
f f kgm T
θ&
f
&&
θ
f
p p 1 3. TS
f f m2
x x s2 4. TR
m m kgm
x&
m
&&
x
m
θ θ rad s 2 5. R
m m kgm 2 R
θ&
m
θ&&
m
p p 1 S
6. R
m m m3
x , x, x x s2 7. T
∫q q q& q& m2 S
x& , x& , x&
∫q q q&
&&
x , &&
x , &&
x
∫q q q&
θ , θ,θ θ rad s 2 8. R
∫q q q& q& m3 S
θ& , θ& , θ&
∫q q q&
θ&& , θ&& , θ&&
∫q q q&
p p kg 9. S
∫q q& m4 S
p
q
p
q&
The 9 different cases correspond to the 9 possible combinations of translational, rotational
and scalar data functions.
QUATTRO RDOF Use: Software Guidelines
The consequent treatment of the correct units must be taken also to the level of analysis.
Establishment of the so-called Index Table (making the data selection for modal parameter
estimation) must support rotational degrees of freedom as (selectable) data entities.
Finally, also on the level of modeshapes, modal participation factors and sensitivities, the
correct units and scale factors must be foreseen and correctly attributed.
The corresponding dimensions of the mode shape residues are given in Table 2:
Case Dimension
1. s
kg
2. rad s
kg m
3. 1
s m2
4. s
kg m
5. rad s
kg m 2
6. 1
s m3
7. s
m2
8. rad s
m3
9. kg
s m4
3. Experimental Modal Analysis
3.1. Data balancing
The main problem encountered with the experimental tools (next to the unit issue as
discussed above), is this of the global treatment of multi-unit data. For example, when
performing a modal analysis, normally all measured FRFs are treated in the parameter
estimation process. The background is that pole values are global system parameters, and
that the redundancy resulting from the global treatment of all data increases the
confidence in the results.
As a consequence, the parameter estimation techniques all use –and minimize- a global
error/cost function, combining the individual measured data. When mixing various data
entities, this may lead to an implicit weighting of those entities as there numerical values
can be largely different. This is not only a problem of rotational versus translational
degrees of freedom, the same problem actually also turns up when treating vibro-acoustic
data (mixing acoustical and vibrational data) or when combining strain and acceleration
data.
QUATTRO RDOF Use: Software Guidelines
In order to avoid that specific data subsets corresponding to the data entities with the
smallest numerical values are dominated in the cost-calculation by the other data subsets,
a balancing of the equations and cost function contributions is to be made.
A similar problem occurs when performing matrix operations on the global data sets, for
example the kernel matrix inversion in FRF substructuring or in inverse load calculations.
These matrix inversions are often ill-conditioned, leading to the use of Singular Value
based pseudo-inverses. When no proper balancing of the different equations is performed,
the data are treated unequally leading to erroneous results.
The balancing factors are then stored and re-applied (inversely) after the execution of the
concerned data operation.
Deliverable D16 discusses various scaling alternatives and gives method
recommendations. An example on this problem of scaling is worked out in D16.
3.2. Other modal analysis aspects
Another problem in experimental modal analysis lies with the representation of the
modeshapes. While normally these modeshapes are represented by animated
deformation shapes, this does not make sense for the rotational degrees of freedom. A
combination of colour plots for the rotations, overlaid on the deformation shapes
representing the translational information, is found to be the best solution.
Experimental modal analysis results are also often used as a basis for further processing,
for example the calculation of forced response predictions, in sensitivity or modification
studies. With forced response calculations, the main requirement is to consistently allow
the generation and application of rotational (moment) load cases.
In sensitivity analysis, this included both the support of rotational DOF modeshape
sensitivities, as well as the support of resonance frequency/damping sensitivities resulting
from stiffness, damping or inertia changes applied in rotational DOFs.
With modification studies, the support of coupling in rotational degrees of freedom must
obviously be supported, but also the “standard” modification solutions (add stiffness, add
inertia..) must include rotational quantities.
QUATTRO RDOF Use: Software Guidelines
4. Transfer Path and Substructuring Analysis
The vibro-acoustic analysis techniques studied in most detail in the project are these of
transfer path analysis and substructuring.
For both techniques, it was concluded that the inclusion of RDOF data could improve the
calculation result significantly. Hence, the tools implementing these methods should
support RDOFS in a consistent way.
4.1. Transfer Path Analysis
In transfer path analysis, there are two steps: (1) calculation of the loads, including the
definition and support of RDOF loads, this means moments, and (2) estimation of the
frequency response functions between the loads and the target responses. The latter
hence implies that FRFs with RDOF inputs must be supported. Also target responses can
be rotational quantities.
In the forward calculation procedure, the loads are multiplied with the propagation FRFs to
yield the partial responses corresponding to the specific load.
For RDOFs, the software hence must support the use of 6 dofs per connection point,
implying both 6-dof loads and transfer functions for 6-dof inputs. Since not all projects will
actually use RDOF information, the choice should be given to use the additional RDOFs or
not. Below, a possible inclusion of the RDOF support in a Transfer Path Analysis tool as it
is available at LMS, is shown.
QUATTRO RDOF Use: Software Guidelines
4.2. Substructuring.
Also in performing FRF-based substructuring (FBS) calculations, it is important to enable
the proper inclusion of RDOF FRF-data.
As such data are not always available, it is recommended to allow selective coupling of the
Degrees of Freedom (e.g. 1 by 1, 3 by 3 or 6 by 6).
QUATTRO RDOF Use: Software Guidelines
An item of particular importance in FBS is that the stiffness characteristics of the coupling
can be defined separately.
These elements were prototyped in the LMS software environment for Frequency
response based substructuring. Some typical menu’s are shown below for the definition of
a substructure and for the definition of the assembly. The RDOF entries can be clearly
seen.
An example of the definition of a flexible coupling matrix is given below.
QUATTRO RDOF Use: Software Guidelines
5. Numerical and Test/FE Analysis
Pure numerical (FE) analysis packages in general do support rotational degree of freedom
quantities adequately. Specific problem occur however when relating experimental and
numerical data. These problems are:
- Orthogonality calculations: The RDOF may play a more significant role than the
translational DOFS.
- Calculation of MAC (Modal Assurance Criterion) functions: this is an inner product
between two modal vectors, scaled by the vector length. Combining RDOF and TDOF
vector information in the same MAC does not make sense, again because of the
scaling issue discussed above. Either the RDOF/TDOFs are first properly scaled, or
the RDOFs and TDOFs are separately treated. The latter solution is preferred, as this
gives a better insight in the separate behaviour of both types of data. The concept of
“groups” of DOFS can easily be used to this purpose, defining separate groups for the
RDOF and TDOF data.
- Similar considerations hold for the calculation of the coordinate-specific correlation
functions like the MACCo. Separate treatment of RDOF and TDOF data is proposed.
On the level of FE model updating, the main implications are with respect to modeshape-
based updating. In principle, the larger the number of equations, the better the updating
result. But even more important than the pure number of different data is the fact that
different data entities and hence different equation types are included in the updating. The
sensitivity information needed for the updating can in principle be calculated from the FE
model itself, so this does not pose practical problems.
A specific, practical, problem was also found on the level of combining solid meshes with
plate/shell components. At the interface between both, the RDOFs of the plate/shell
cannot be coupled directly to the solid. Workarounds have been developed by adding
extra elements (additional beams or rib, or penetrating the plate in the solid…).
6. Residual compensation
When the behaviour of a coupled structure needs to be estimated from the modal
characteristics of the two substructures, the information is often obtained from
experimental data. “Experimental” means that we have only the experimental FRFs
between some DOFs and the mode shapes that have been obtained through a modal
analysis method, such as Time Domain MDOF for example. These mode shapes have
not a very important number of DOFs (it is limited to the number of DOFs of the
wireframe), which explains the spatial reduction, and they belong to a limited frequency
band, which accounts for the frequency incompleteness. These two incompletenesses
lead often to a wrong dynamic behaviour of the coupled structure because high frequency
mode shapes of the substructures can play an important part. The aim of the residual
modes is to compensate for these omitted modes by calculating new modes from the
difference between the FRFs obtained from the tests and the FRFs synthesized with only
the modes taken into account.
QUATTRO RDOF Use: Software Guidelines
An approximation of the residual modes effect is a quadratic function using a taylor
expansion as shown in equation [1] and figure 8. A compensation procedure is developed
by LMS to compensate for the truncation effects by using the direct FRFs. In case of
experimental data these driving point FRFs are available. In case of numerical simulations
the truncation error can be estimated using a direct FRF evaluation at 2 frequencies and
estimating the quadratic function.
[H (ω)] ≈ [H
C R
ωt ] [ ] [ ]
(ω) + Hω0t + ω2 Hω1 t [1]
exact FRF static
superposion compensation
modal dynamic
FRF compensation
A procedure is developed to convert the residual quadratic error into residual modes using
a singular value decomposition of the residual FRF matrix, as shown in equations [2] – [5].
[H C
] [ ] [ ]
(ω ) ≈ H ωRt (ω ) + H ω0t + ω 2 H ω1 t [ ] [2]
[H ]+ [H ].ω
0
ω
1
ωt
2
= [Ψh ] ( [ ω ]− ω ) [Ψ ] ≈ [Ψ ] [ ω ][Ψ ]+ [Ψ ] [ ω ][Ψ ].ω
\ 2
r\
2 −1 T
h h
\ −2
r \
T
h h
\ −4
r \
T
h
2
[3]
[H ] = [Ψ ] [ ω ][Ψ ]
0
ω h
\ −2
r \ h
T
[H ] = [Ψ ] [ ω ][Ψ ]
[4]
1 \ −4 T
ωt h r \ h
[H ] = [U ][Σ ][U ]
0
ωt
0 0 0 T
⇒
[ ] −1
ω r \ estim = diag U 1 H 0
\ 2
([ ] [ ])
[H ] = [U ][Σ ][U ] [ ] [ ][ ]
[5]
1
ωt
1 1 1 T
[Ψ ]estim = U 0 Σ0 . \ω r \
Besides the new developed SVD-based technique, several other methods for residual
compensation exist, in which extra mass (M), stiffness (K) and damping (C) is added at the
coupling points. All techniques calculate an extra set of residual modes outside the
frequency band of interest. The basic procedure is as follows :
First of all, one has to choose the two frequencies :
• the static frequency
• the dynamic frequency
QUATTRO RDOF Use: Software Guidelines
Then one has to decide if we want a single frequency method or a least squares method
(frequency band method).
Finally one has to choose the technique :
1.a. SVD Static Only
1.b. SVD Static Log (this method is the same as the SVD Static method, but based
on the log difference between the FRFs)
2. SVD Dynamic
3.a. MK Method
3.b. MKsym Method (this method compelles the estimation of the two matrices to
be symmetric)
4.a. MKC Method
4.b. MKCsym Method
There are therefore 7 different methods, with two choices (least squares or not), so when
the static and dynamic frequencies are fixed, 14 choices for the method are available.
7. Conclusion
As a result of the work in Task 5.3, a complete understanding is now available of the
various consequences of using RDOF information in engineering software tools for
CAT/CAE. The implementation of these recommendations will be the basis for the
exploitation of the project results by the software tools vendors.
