See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/267952994

MODAL AND FREQUENCY BASED

SUBSTRUCTURING USING ROTATIONAL DOF
CONSIDERATIONS

Article

CITATIONS READS

5 54

3 authors, including:

Christopher Cook Chipman

Diversified Technologies, Inc
5 PUBLICATIONS 34 CITATIONS

SEE PROFILE

All content following this page was uploaded by Christopher Cook Chipman on 12 February 2015.

The user has requested enhancement of the downloaded file.

 MODAL AND FREQUENCY BASED SUBSTRUCTURING
USING ROTATIONAL DOF CONSIDERATIONS

Aaron Williams
Christopher Chipman
Dr. Peter Avitabile
Structural Dynamics and Acoustic Systems Laboratory
University of Massachusetts Lowell
Lowell, Massachusetts

ABSTRACT

System models are generated from component dynamic information using modal based and frequency based assembly
techniques. Often times the lack of residual information and rotational information causes errors in these approaches.
Several different system modeling approaches are used and compared for a simple structure. Both modal based and
frequency based approaches are utilized. Inclusion of residual information in the system model is provided for improved
system model results. Estimation of rotational effects are also derived and included in the system model using several
approaches. The results of all the techniques are compared and discussed.

INTRODUCTION

System models are often developed from a variety of different sources. Modal based and FRF based models are commonly
used. When experimental data is included as part of the component description, often rotational DOF are not available due
to measurement limitations. Many times approximations of these difficult to obtain measurements are obtained using
equivalent translational DOF or using expansion methods to estimate the RDOF.

In this paper, several different approaches are used to describe a system model developed from two components which
have only translational measured DOF but require rotational DOF for the development of the system model from
component representation. Both modal models and frequency based models are used for the system description and results
are compared.

THEORY AND PROPOSED METHOD

In order to present the results of the studies performed herein, there are several basic fundamental theoretical approaches
that need to be summarized. These relate to model reduction and expansion, system modeling approaches and frequency
based substructuring techniques. Each of these methods is only summarized here; details of the techniques are contained in
the respective references.

Model Reduction:

Model reduction is necessary in order to develop expansion approaches for modal data as well as frequency response
function expansion for rotational degrees of freedom. These techniques have been presented in earlier work cited in the
references; only summarizing equations are presented below. Several model reduction methods have commonly been used
for expansion of measured data. Four common methods are Guyan [1], Dynamic Condensation [2], SEREP [3], and a
Hybrid method [4]. In these methods, the relationship between the full set of degrees of freedom and a reduced set of
degrees of freedom can be written as:

{X n }=[T]{X a } (1)
 All of these methods require the formation of a transformation matrix that can project the full mass and stiffness matrices
to a smaller size. The reduced matrices can be formulated as:

[M a ] = [T]T [M n ][T ] (2)

[K a ] = [T ]T [K n ][T] (3)

The Guyan reduction process [1] forms the transformation matrix as:

[I]
[TG ] = ⎡⎢ ⎤
⎥
(4)
⎣− [K dd ] [K da ]⎦
−1

The Dynamic Condensation process [2] (which is a variation of the Guyan process using a shifted eigenvalue concept)
forms the transformation matrix (and Dynamic Matrix [D]) as:

[I]
[TD ] = ⎡⎢ ⎤
⎥ ; [D] = [[K ] − f [M]]
⎣− [D dd ] [D da ]⎦
−1 (5)

The System Equivalent Reduction Expansion Process (SEREP) [3] produces reduced matrices for mass and stiffness that
yield the exact frequencies and mode shapes as those obtained from the eigensolution of the full size matrix. The SEREP
transformation is formed as:

[TU ] = [U n ][U a ]g (6)

The Hybrid method [4] utilizes the accuracy of the SEREP method and seeds the reduced matrices with reduced Guyan
matrices to insure that the resultant reduced matrices are fully ranked for all cases. The Hybrid method transformation
matrix is:

[TH ] = [TG ] + [[TU ] − [TG ]][[ U a ][U a ]T [TU ]T [M n ][TU ]] (7)

Equation (1) is used for expansion of measured data, [E], and is written as:

[E n ] = [T][E a ] (8)

The transformation matrix, [T], can be any one of the matrices used for reduction described above.

Modal Model for System Model Development :

Modal models [5] are commonly used to develop system models where components are each tested independent of the
assembled system. The basic system modeling equation for attaching Component A to Component B (using a stacked
modal vector arrangement for the modal transformation) is given as:

⎡ ⎡ ⎡O ⎤ ⎤ ⎤⎧ ⎫ ⎡ ⎡ ⎡O ⎤ ⎤ ⎤⎧ ⎫
⎢⎢⎢
⎢⎢⎢ M A ⎥
⎥ ⎥
⎥ ⎥⎪ A
{ }
⎥ ⎪ &p&
⎪ ⎢⎢⎢
⎪ ⎢⎢⎢ K A ⎥
⎥
⎥
⎥ { }
⎥⎪ A
⎥⎪ p
⎪
⎪ (9)
⎢ ⎢ ⎢⎣ O⎥⎦ ⎥ ⎥ ⎪⎪ ⎪⎪ ⎢ ⎢ ⎢⎣ O⎥⎦ ⎥ ⎥ ⎪⎪ ⎪⎪
⎥ + [U ] [ΔM ][U ]⎥ ⎨ ⎬ + ⎢⎢ ⎥ + [U ] [ΔK ][U ]⎥ ⎨ ⎬ = {0}
T T
⎢⎢
⎢⎢ ⎡O ⎤⎥ ⎥⎪ ⎪ ⎢⎢ ⎡O ⎤⎥ ⎥⎪ ⎪
⎢⎢
⎢⎢
⎢
⎢ MB ⎥⎥
⎥⎥ ⎥⎪
{ }
⎥ ⎪ &p& B ⎪ ⎢⎢
⎪ ⎢⎢
⎢
⎢ KB ⎥⎥
⎥⎥ { }
⎥⎪ pB
⎥⎪
⎪
⎪
⎣⎢ ⎢⎣ ⎢⎣ O⎥⎦ ⎥⎦ ⎦⎥ ⎪⎩ ⎪⎭ ⎣⎢ ⎢⎣ ⎢⎣ O⎥⎦ ⎥⎦ ⎦⎥ ⎪⎩ ⎪⎭
Impedance Model for System Model Development :

Impedance models [6,7] are commonly used to develop system models where components are each tested independent of
the assembled system. In the case of this model, a free-free component can be attached to another component using only
the attachment DOF measured FRFs. This technique has been widely used and the basic equation describing the response
of the system can be defined as

h Cij = h Aij − ⎣H A ⎦iS ([H A ]SS + [H B ]SS )−1 {H A }Sj (10)

The form of this equation specifically describes the transfer functions of the outputs on Component A due to inputs on
Component A as influenced by the dynamic effects of Component B. Variations on this one equation cover the range of
possibilities that can exist in the development of a system model.

In order to assemble two components together where RDOF are necessary, then the measured frequency response functions
must include RDOF. While several variations are possible [8], a commonly used expansion approach utilizing SEREP for
mode expansion and Dynamic Condensation for residual expansion allows for the development of the necessary RDOF.
This is referred to as the SEREP/DRE method and is given as:

⎧ ⎧ h ij ⎫ ⎫ ⎧⎧⎪ M q k u ik u jk ⎫⎪⎫ ⎧ Upper Residual ⎫

⎪⎨ ⎬⎪ ⎪⎨ ∑ + *⎬⎪ ⎪ ⎪
⎪⎪ ⎩Tdof ⎭ ⎪⎪ ⎪⎪⎪⎩k =1 j ω − p ⎪⎭⎪ ⎪⎪ for Translatio n DOF ⎪⎪ (11)
k
⎪
⎨− − − − ⎬ = ⎨− − − − − − − − − −⎬ + ⎨− − − − − − − − − − − − − − − − ⎬
⎪⎧ h ⎫⎪ ⎪ SEREP Expansion ⎪ ⎪ DynamicExpansion ⎪
⎪⎨ ij ⎬⎪ ⎪ ⎪ ⎪ ⎪
⎩⎪⎩Rdof ⎭⎭⎪ ⎪⎩ for RotaryDOF ⎪⎭ ⎪⎩ for RotaryResidualDOF ⎪⎭

where the SEREP expansion is given by

[ ]
[T U ] = [U n ] U ga (12)
and the Dynamic Expansion is given by
{UR Rdof } = [D dd
−1
D da ]{UR Tdof } (13)

The analytical tools described above are used in the development of the system models described next.

COMPONENT MODELS DEVELOPED

For this study, a simple two beam system model will be used for the generation of experimental modal models and
necessary frequency response functions for system model development. The basic beam component used is shown in
Figure 1; this beam is used to form a system where the two beams are bolted together.

Figure 1 – Component Beam Geometric Characteristics

The beam is made of aluminum and is modeled with 15 nodes (15 translational and 15 rotational DOF). The beam was
tested using a multiple reference impact technique to obtain measured frequency response functions. Typical modal
parameter extraction techniques were employed to obtain the modal data for the beam. All data was collected and
processed using the LMS CADA-X suite of acquisition and reduction software packages [9]; additional processing was
also performed using MATLAB [10].

The instrumented structure is shown in Figure 2 along with the schematic of the test setup.

Figure 2 – Typical Beam Measurement Setup

In order to obtain highly accurate measurements, a pendulum impactor was used in lieu of the traditional impact hammer
approach and is shown in Figure 3. This was necessary in order to obtain FRF accuracy needed for both resonances and
anti-resonant regions of the frequency response measurements.

Figure 3 – Impact Pendulum Measurement Setup

A comparison of the analytical frequencies from the finite element model and the tests frequencies from the experimental
modal test are shown in Table 1. The lower order modes correlate well. The shapes are shown in Figure 4.
 Table 1 – Comparison of FEM and Tested Frequencies for Beam Component
FEM Test % Diff
127.58 126.60 0.77
361.76 353.14 2.44
717.62 684.17 4.89
1184.35 1092.89 8.37
1744.61 1545.75 12.86

Figure 4 – Lower Order Mode Shapes of the Beam Component

REFERENCE SYSTEM MODEL

The reference model used for this study was developed as an assembled finite element model and also tested in the final
assembled configuration for comparison of solutions. The resulting frequencies and frequency response functions are
included in the comparison plots and tables as necessary (for comparison to results of the cases studied). An analytical
model for the two beams tied together was developed and the assembled structure was tested for reference. The resulting
system frequencies are shown in Table 2.
Table 2 –Tested Frequencies for Two Beam Assembled System
FEM TEST % Diff
31.37 30.75 2.02
98.33 97.00 1.37
172.48 166.50 3.59
322.78 313.50 2.96
433.32 413.50 4.79
674.46 638.00 5.71

CASES STUDIED

The component models were used to develop system models using different approaches. Each of the various models is
described as follows.

Case 1 involves modal models with RDOF expansion for system model development
Case 2 involves FBS models with only TDOF measurements
Case 3 involves FBS with RDOF expansion for FRF synthesis
Case 1 - Modal Models with RDOF Expansion

Modal Case 1a – Experimental Component Assembly – 2 Analytical RBM + 5 experimental modes

This case involved the use of actual test data, which include only TDOF, as shown in Figure 5.

Figure 5 – Only TDOF obtained from Beam Test

The 5 test modes extracted from the beam were expanded to include both TDOF and RDOF. Appended to these test modes
were two analytical rigid body modes (RBM); this hybrid mode set was used to create a modal model of the two beam
system, tying connection DOF with a spring stiffness of 1E6 lbf/in. A MAC comparing this modal model to the analytical
component assembly is displayed in Figure 6. Good correlation is observed up to mode 14, where truncation affects the
modal solution.

Figure 6 – MAC – Two Beam System Model – 5 Test Modes and 2 Analytical RBM

Modal Case 1b – Experimental Component Assembly – 2 Analytical RBM + 5 experimental modes + 6 Analytical Modes

In order to improve the solution, more modes were required. Six additional higher frequency analytical modes were
appended to the component modes. This mode set includes 2 analytical RBM, 5 test modes expanded to include RDOF,
and 6 higher order analytical modes. A MAC comparing the assembled modal model created with this new hybrid mode
set to the reference component assembly is shown in Figure 7. The correlation is greatly improved with the addition of
these higher order modes to account for residual effects.
 Figure 7 – MAC – Two Beam System Model – 5 Test Modes, 2 Analytical RBM and 6 High Order Analytical Modes

Modal Case 1c – Experimental Component Assembly – Full Set of 30 Modes

Appending 23 additional higher order modes created a hybrid mode set of 30 modes (which consists of as many modes as
DOF). A MAC comparing the modal model created using this full mode set to analytical is shown in Figure 8. Good
correlation across all 30 modes is observed. The frequencies from each modal model are compared to the reference solution
in Table 3.

Figure 8 - MAC – Two Beam System Model – Full 30 Modes

 Table 3 – Frequency Comparison between Modal Models
Case 1a Case 1b Case 1c
Test Test
Analytical Test +2 RBM +2 RBM
+2 RBM +6 Higher +23 Higher
Analytical Analytical
31.37 31.74 30.76 30.73
98.32 97.53 97.53 97.53
172.47 173.51 169.22 169.13
322.72 315.03 315.01 315.00
433.25 430.88 420.42 420.21
674.22 643.05 642.97 642.95
813.10 793.86 774.45 774.12
1142.92 1055.62 1055.43 1055.39
1305.87 1247.85 1212.31 1211.87
1712.25 1519.16 1518.83 1518.79
1897.43 1960.36 1745.68 1745.01

In looking at the results in Table 3, it is important to recall that only translational DOF are measured for the beam. The
rotational DOF are computed from an expansion using the simple beam model used to represent the component. Because
only a simple beam is used to describe the rotational effects, there is no inclusion of the flange flexibility – it is modeled as
a rigid mass. This may have a significant effect on higher modes of the system and is discussed later in this paper.

Case 2 - Frequency Based Models

Frequency response data collected from a single beam is used to approximate a rigidly coupled two-beam system. Purely
test based FBS cases including the use of a force couple on the beam flange to approximate the bolted connection were
performed along with cases involving the synthesis of RDOF FRFs from a combined set of analytical and expanded test
modes for the modal beam. All FRF comparisons between FBS and Experimental data are given for translational point 4
drive-point measurements on the physical beam as shown in Figure 9.

Figure 9 - Drive Point 4

Case 2a – Test Based FBS Using Only TDOF

The most straightforward FBS case utilized x-axis FRFs at points 1 and 2 on the end flange as a force couple to resist
rotary motion about the z axis. FRFs were also acquired for the y-axis at both points 1 and 2 and were coupled to resist y-
direction translation. This case is the most direct method as the FRFs required were all translation and came directly from
the impact test. The Case 2a connection is shown along with FRF results compared to the experimental system test
measurements in Figure 10. Results show good correlation especially in the low frequency range.
 Figure 10 - Comparison of Experimental and Case 2a FBS Results

Case 2b – Test Based FBS Using Only TDOF – Center Flange Connection

A second case was examined including the force couple in the Tx direction along with a single Ty rigid couple estimation
at point 3 to equate the translation at the center of the beams. A center-flange y-axis drive-point measurement was used
involving an impact at point 1 and a measurement at point 2. This case was used in order to gain insight into the sensitivity
of the two-beam system response to the location of the y-axis coupling point and respective drive-point measurement
characteristics. The Case 2b connection is illustrated along with FRF results compared to the experimental system test
measurements in Figure 11.

Figure 11 - Comparison of Experimental and Case 2b FBS Results

Table 4 shows a peak pick comparison for the experimental system test and the FBS cases utilizing test acquired translational
DOF only measurements. Results for these cases are comparable to the tested system frequencies.
 Table 4 - FBS Results Using Only TDOF FRF Measurements

Test Case 2a Case 2b

30.75 33.94 33.94
97.00 97.40 97.40
166.50 166.11 166.11
313.50 314.89 314.89
413.50 412.92 412.92
638.00 662.59 647.36
756.50 756.22 756.22

Case 3 – FBS Models with RDOF Expansion

Three expanded mode shape cases were used to perform FBS using translational and rotational DOF ties. The solution
scheme involved the synthesis of translational and rotational transfer functions in order to model the two-beam system with
a rigid connection at the beam end-node. These cases involved only y-axis translation and Rz-axis rotational DOF, no x-
direction connections are used for Case 3 FBS solutions. (Again it is important to note here that only a very simple beam is
used for the system expansion equation; the flange is considered as a rigid mass with no flexibility.)

Case 3a – FBS – 2 Analytical RBM + 5 experimental modes

The first expanded FBS case included only the 2 analytical rigid body modes and 5 flexible modes from test. The resulting
FRF comparison along with connection illustration is shown in Figure 12. The results are fairly good especially in the
lower frequency range where truncation effects are not as serious.

Figure 12 - Comparison of Experimental and Case 3a FBS Results

Case 3b – FBS – 2 Analytical RBM + 5 experimental modes + 6 Analytical Modes

A second expanded mode shape set consisted of the two rigid body modes and 5 test modes along with 6 higher analytical
modes. FRF comparison is shown in Figure 13. The results are slightly improved from the previous case.
 Figure 13 - Comparison of Experimental and Case 3b FBS Results

Case 3c – FBS – Full Set of 30 Modes

The third expanded mode shape set included rigid body modes, 5flexible modes and 23 higher order flexible modes for a
total of 30 modes used for FRF synthesis. Resulting FBS system response comparison is given in Figure 14. Frequency
results comparison for the FBS cases involving a rotary DOF tie are shown with experimental test frequencies in Table 5.

Figure 14 - Comparison of Experimental and Case 3c FBS Results

Table 5 - FBS Results for Expanded Mode Shape Cases

Test Case 3a Case 3b Case 3c

30.75 33.94 33.94 33.94
97.00 97.40 97.40 97.40
166.50 183.75 178.68 176.98
313.50 314.89 314.89 314.89
413.50 457.86 440.94 440.94
638.00 643.98 643.98 643.98
756.50 845.32 811.48 811.48
As further comparison the two test-based FBS cases have been overlaid with the experimental results in Figure 15. The
Case 3 expanded mode set FBS results are compared in Figure 16 to show the effects of truncation when higher order
analytical modes are not included.

Figure 15 - Comparison of Experimental and Test Based FBS Results

Figure 16 - Comparison of FRF Results for Expanded Mode Set FBS Cases
Table 6 is included for further comparison of the results for both Modal and FBS techniques involving test data expanded
to include rotary DOF ties at beam endpoints. Generally, the results are reasonably good especially considering the lower
order modes. The results are also considered good realizing that only simple beam bending expansion functions were used
to estimate the RDOF FRFs.
 Table 6 – Comparison between Case 1 Modal Models and Case 3 FBS
Test + 2 RBM + 6 Test + 2 RBM + 23
2-Beam
Test + 2 RBM Higher Order Higher Order
System
Analytical Analytical
Modal FBS Modal FBS Modal FBS
Test
Case 1a Case 3a Case 1b Case 3b Case 1c Case 3c
30.75 31.74 33.94 30.76 33.94 30.73 33.94
97.00 97.53 97.40 97.53 97.40 97.53 97.40
166.50 173.51 183.75 169.22 178.68 169.13 176.98
313.50 315.03 314.89 315.01 314.89 315.00 314.89
413.50 430.88 457.86 420.42 440.94 420.21 440.94
638.00 643.05 643.98 642.97 643.98 642.95 643.98
756.50 793.86 845.32 774.45 811.48 774.12 811.48

OBSERVATIONS

The results of the Modal Solution (Case 1) and FBS Solution (Case 3) generally follow the same trend in terms of the
results. The main difference is attributed to FBS enforcing a constraint between the two beams and the Modal Solution
used a very stiff spring to tie the beams together.

But when comparing the results to the FBS Solution for Case 2, the accuracy obtained is not solely due to truncation
effects. In the models of both Case 1 and Case 3, the effects of rotational DOF were provided by an expansion process.
However, the expansion was performed by using a very simple beam representation of the system. This beam model did
not attempt to model the flange flexibility in any detail; the flange was modeled as a lumped rigid mass in the model with
only transverse and rotary inertial properties. Subsequent to these studies, a more detailed model of the beam component
was made and the flange is seen to have some significant flexibility in the higher frequency ranges (with one flange
flexible mode shown in Figure 17). This flexibility can not be handled by the simple beam expansion function and shows
that additional care and consideration is needed in order to provide a more accurate model.

It is important to note, however, that the models of Case 2 do produce more accurate results mainly due to the fact that the
additional measurements used as coupling points on the flange do in fact properly characterize the flange flexibility for the
FBS solution approach. (If additional time was available, the extension of the modal solution to couple the two beams
together at these same flange points would most likely show improved results comparable to the FBS solution.)

Figure 17 – Single Beam Local Flange Mode at Approximately 1660Hz

 CONCLUSIONS

System models using Modal Approaches and Frequency Approaches were presented which involved both translation and
rotation DOF. Different methods for development of the models which required rotational system connections were
presented. The results from all methods were compared and show similar trends in the accuracy of the final system model
results. The differences were discussed to highlight some areas that need to be carefully considered when developing these
types of models.

REFERENCES

1 Guyan, R.J., "Reduction of Stiffness and Mass Matrices", AIAA Journal, Vol. 3, No. 2, 1965
2 Kidder, RL, "Reduction of Structure Frequency Equations", AIAA, Vol 11, No 6, June 1973, p. 892
3 O'Callahan, J.C., Avitabile, P., Riemer, R., "System Equivalent Reduction Expansion Process", Seventh International
Modal Analysis Conference, Las Vegas, Nevada, February 1989
4 Kammer, D. C., "A Hybrid Approach to Test-Analysis-Model Development for Large Space Structures," Journal of
Vibration and Acoustics, Vol. 113, No. 3, July 1991, pp. 325-332
5 Avitabile, P., “Twenty Years of Structural Dynamic Modification – A Review”, Proceedings of the Twentieth
International Modal Analysis Conference, Los Angeles, California, Feb 2002
6 Avitabile, P., Piergentili, F., Lown, K., “Application of Hybrid Modeling Techniques for Computer Related
Equipment”, Fifteenth International Modal Analysis Conference, Orlando, Florida, Feb 1997
7 Avitabile, P., A Comparison of Some System Modeling Approaches, The Shock and Vibration Digest, Vol 33,
Number 4, July 2001, pp 281-291
8 Piergentili, F., Rotational Degree of Freedom Estimation of Frequency Response Functions for Substructured
Experimental Components, PhD Dissertation, Mechanical Engineering Department, University of Massachusetts
Lowell, 1999
9 CADA-X Modal Software, Leuven Measurement Systems, Leuven, Belgium
10 MATLAB 6.5 – The MathWorks, Natick, Massachusetts

View publication stats