Mechanical Systems and Signal Processing 31 (2012) 5666

Contents lists available at SciVerse ScienceDirect

Mechanical Systems and Signal Processing

journal homepage: www.elsevier.com/locate/ymssp

Advancements in hybrid dynamic models combining experimental and nite element substructures
R.L. Mayes n, M.R. Ross 1
Sandia National Laboratories, P.O. Box 5800, MS 0557, Albuquerque, NM 87185-0557, USA

a r t i c l e in f o
Article history: Received 11 May 2011 Received in revised form 11 February 2012 Accepted 21 February 2012 Available online 28 April 2012 Keywords: Component Mode Synthesis Substructuring Experimental Dynamic Substructuring

abstract
This paper presents very practical enhancements to the transmission simulator method (TSM); also known as the Modal Constraints for Fixtures and Subsystems (MCFS). The enhancements allow this method to be implemented directly in nite element software, instead of having to extract the reduced nite element model from its software and implement the substructure coupling in another code. The transmission simulator method is useful for coupling substructures where one substructure is derived experimentally and the other is generated from a nite element model. This approach uses a exible xture in the experimental substructure to improve the modal basis of the substructure; thus, providing a higher quality substructure. The exible xture substructure needs to be removed (decoupled) from the experimental substructure to obtain the true system characteristics. A modied method for this removal and coupling of the experimental and analytical substructures is provided. An additional improvement guarantees that the experimental substructure matrices are positive denite, a requirement for many nite element codes. Guidelines for designing robust transmission simulator hardware are provided. The concepts are applied to two sample cases. The rst case consists of a cylinder connected by eight bolts to a plate with a beam. The second example is an outer shell structure that is connected through a bolted ange to a complex internal payload structure. & 2012 Elsevier Ltd. All rights reserved.

1. Introduction A complex full system model can be a synthesis of individual components (substructures). Typically, these components are designed and analyzed by different groups of engineers, and often they are in different organizations if not companies. However, the development of one component requires knowledge of the characteristics of the other components to predict the response of the full system. The engineering team of one component would prefer not to model or expend additional computational resources on other substructures. Several analytical component mode synthesis (CMS) techniques have been developed to address these issues [18], but these methods are difcult to implement in experimental substructures. Possibly, an experimental substructure may provide a more accurate model as well as a more economical one. This paper enhances the current body of knowledge on the CMS techniques for coupling experimental and analytical models. It employees a technique developed by Allen, Mayes and Bergman termed Modal Constraints for Fixtures and
n

Corresponding author. Tel.: 1 505 844 5324. E-mail address: rlmayes@sandia.gov (R.L. Mayes). 1 Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energys National Nuclear Security Administration under contract DE-AC04-94AL85000. 0888-3270/$ - see front matter & 2012 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.ymssp.2012.02.010

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

57

Subsystems (MCFS) with the use of a exible xture named the transmission simulator (TS) [9]. Hereinafter, this will be called the Transmission Simulator method (TSM). The basic premise of their concept is to use an additional substructure (transmission simulator) at the interface of two components. This additional substructure allows for the joining of two components with multiple connection points. It is connected to the experimental substructure to improve the modal basis and reduce the required number of measurement points. In addition to the experimental TS, an analytical model of the TS was attached to the nite element (FE) substructure model. The two substructures were coupled and then both transmission simulators were subtracted. The modied TSM addressed here does not require the analytical TS be attached to the FE substructure model. The experimental TS is rst subtracted from the experimental substructure. Then, a constraint equation is used to couple the resulting experimental substructure to the FE substructure model. With this enhancement, the coupling can be performed within the FE code instead of extracting the FE substructure and combining it with the experimental substructure in a separate code. The TSM removes the difcult issues commonly found with experimental substructures: measuring connection rotational responses and moments, and measuring constraint modes or residual attachment modes with both translations and rotations at multiple connection points. It is important that the modes of the experimental substructure system form an adequate basis for the motion of the coupled system, as is the case with any modal substructuring technique. The TSM method also provides additional advantages when combining analytical and experimental substructures. The TSM captures the damping and stiffness of not only the experimental substructure, but also the joint between the experimental and FE substructure models. In addition, the TSM provides relief from experimental errors in the mode shape measurements. A number of researches have investigated the combining of analytical and experimental substructures. The classical approach of Craig-Bampton that uses xed interface and constraint modes [2] is difcult to implement experimentally, because the perfectly xed boundary condition cannot be achieved. Another popular set of basis vectors addressed in the literature [4,6] is a set of modes comprised of the rigid body modes, kept free interface modes and residual exibility attachment and/or inertia relief modes. The free interface modes are easy to measure and the rigid body modes can be either measured or calculated. Often these are not accurate enough without the residual exibilities, because they do not capture the strain near the joint. Martinez et al. used free modes with residual exibility [10]. Yet, the residual exibilities between all the attachment degrees of freedom (DOF) are not easily measured, especially the rotational/moment terms. This is signicantly more difcult for many attachment DOF. An attractive approach to improve the modal basis vectors experimentally is to attach masses at the coupling interface of the experimental substructure [11]. Unfortunately, if there are several connection points, this can become infeasible, for there will be a need for six sensors per connection point to determine all the displacements and rotations. Corus et al. [12] provided a method for structural dynamic modication with similarities to the TS method. They generated a coarse nite element model of a local portion of a structure to be modied and added to it a model of the conceived modication. Then an experimental modal model of the original structure was measured and extracted. Translation measurement points on the structure that coincided with DOF on the coarse model were thus expanded to the translations and rotations connected to the new modication model using the mode shapes of the coarse nite element model. The transmission simulator method can be viewed as an extension to the approach where rigid masses are attached to the experimental substructure [13]. However, it does not require as many sensors as physical connection DOF. This paper adds to Allen et al. modal constraint method [9] with the following: it provides a computationally efcient implementation appropriate for use in FE codes, design recommendations for the transmission simulator are offered, and a new diagnostic is addressed to eliminate indenite matrices. The computationally efcient implementation uses multi-point constraints (MPCs) to connect the experimental substructure into the nite element substructure. This is more convenient for the nite element analyst. Previously, substructure models of the nite element substructure and experimental substructure had to use an additional computational code to perform the coupling as well as the xture subtraction. This may not be convenient if the forces are widely distributed over thousands of DOF in the FE model. The calculation of the right hand side moving phi transpose times the force vector to another code can be tedious or even prohibitive. For these cases, it would be easier to bring the relatively small modal model of the experiment into the FE analysis code and connect it with constraints to the nite element model. Finally, the TSM also allows one to analyze the nite element model with different types of solutions (transient, non-linear transient, modal, etc.). The paper is organized as follows. Section 2 provides the theory for the new MPC approach. This section also discusses a methodology to avoid indenite matrices that can happen with the TSM. Section 3 provides verication examples of the new MPC approach. Section 4 elaborates on the transmission simulator design and desired characteristics. The appendix provides the original formulation for the Modal Constraint method so that the paper can stand alone for completeness. 2. Theory This paper presents a very practical enhancement to the transmission simulator method (TSM); also known as the Modal Constraints for Fixtures and Subsystems (MCFS). The original theory is presented in Appendix A. The derivation begins with the coupling of a nite element substructure to an experimental substructure. The derivation assumes that the Transmission Simulator has already been removed from the experimental substructure in a previous step. The removal of the Transmission Simulator from the experimental substructure will be discussed after the coupling of the nite element substructure to the experimental substructure. It is important to note that many nodes of the analytical Transmission

58

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

Simulator will be perfectly coincident with many nodes of the nite element substructure. However, the analytical Transmission Simulator is not included as part of the FE substructure as previously proposed [9]. 2.1. Coupling experimental substructure to the FE substructure The derivation begins with the coupling of the experimental substructure to the nite element substructure. Here the original theory is modied producing the following uncoupled equations of motion for two substructures that can be cast entirely in a nite element modeling code as 3( " #( ) " #( ) 2 T ) FE K FE 0 0 xFE I FE f FE x 0 M FE 4 5 , 1 expTS 0 UT x2 0 qexpTS f expTS q 0 I expTS expTS expTS where the FE subscript is for the nite element model, exp-TS is the subscript for the experimental substructure with the TS removed (the removal is explained in Section 2.2), x are physical displacement DOFs, q are generalized modal displacements of the experimental substructure, K is the stiffness matrix, M is the mass matrix, x2 is the experimental substructures diagonal stiffness matrix of frequency in (rad/sec) squared, IexpTS is the experimental substructures diagonal mass matrix that is the identity matrix with mass normalization of the modes, IFE is the identity matrix the size of T the nite element model, UexpTS is the transpose of the experimental substructures mode shapes normalized to unity modal mass, and f are forces. In this paper, lower case bold variables and/or {} denote vectors and upper case bold variables and/or [] denote matrices. In the previous implementation, described in Appendix A, the substructures included the Transmission Simulator in both the FE and the experimental substructures, but here this is not the case. An advantage in this approach is that the TS does not need to be attached to the FE substructure as done in previous versions. A FE model of the transmission simulator is still retained and its mode shapes are used later; dene H as mode shapes of the TS alone. The substructures in the equation of motion, Eq. (1), are coupled with the constraint
coin xcoin FE xexpTS 0,

where the superscript coin denotes DOFs for coincident nodes between the TS and the FE model. A very subtle but important point here is that the FE model of the TS simulator shall be meshed to have several coincident nodes with the FE model in the region where the TS and FE substructure occupy the same space. So even though this improvement does not require the TS to be connected to the FE substructure, it has to have nodes located coincident with the nodes of the analytical TS. There are also DOFs in the TS analytical model at the measurement locations. The signature equation of the TS method is provided by premultiplying Eq. (2) by the pseudo-inverse (y) of the TS mode shape matrix, giving h iy h iy coin Hcoin xcoin xcoin 3 TS FE HTS expTS 0: This represents a least squares solution to the constraint equation, Eq. (2). In [9] this approach was found to give dramatically better results than attempting to use a modal lter to determine the connection point responses. Consider the motion of the second term in Eq. (2) approximated as
coin xcoin expTS UexpTS qexpTS :

Substituting Eq. (4) into the last term of Eq. (3) gives h iy h iy coin Hcoin xcoin Ucoin TS expTS qexpTS 0: FE HTS Now, consider a matrix P such that 2 3 " coin # Ucoin HTS expTS 4 meas 5, meas P 

HTS

UexpTS

where the superscript meas denotes the measurement DOF in the modal test. Therefore,
coin Hcoin TS P  UexpTS , and meas Hmeas TS P  UexpTS :

7 8

P can be calculated as h iy meas y meas Ucoin UexpTS , P Hcoin TS expTS HTS

showing the equivalence of the second and third terms. The experiment only provides mode shapes at the measured DOF, so substituting the third term in Eq. (9) for the second term in Eq. (5) yields h
meas Hcoin xcoin Umeas TS expTS qexpTS 0: FE HTS

iy

10

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

59

Eq. (10) provides the multi-point constraints (MPC) for the FE code to couple the experimental substructure to the FE substructure nodes that are coincident with the TS substructure. Physically, Eq. 10 connects all the measurement points of the experimental substructure on the TS (in the second term) with all the nodes in the FE model where the TS would have been welded in by the previous method [9]. The rst term in Eq.10 includes all the nodes where the TS would physically overlap with the FE model, which is a much larger number of nodes, typically, than the measured nodes. In the FE code the experimental model can be implemented as a group of single DOF freedom systems. Each of the qexp-TS DOF are assigned modal mass of one and connected to ground by a spring with a stiffness value coming from the appropriate diagonal term of the stiffness matrix x2 expTS , which will come from the experimental substructure analysis in section 2.2 The MPC of Eq. 10 constrains the single degree of freedom systems, representing the experimental substructure, to the FE substructure nodes with the coin superscript. The resulting coupled equations of motion can be solved to provide new natural frequencies and mode shapes for the hybrid FE/experimental system. 2.2. Removal/Uncoupling of the TS from the experimental substructure How does one obtain the stiffness matrix x2 expTS and the mode shape matrix Uexp-TS, which are inputs to Eq. (1)? A modal test is performed on the experimental hardware which includes the TS so that modal frequencies and shapes scaled to unity modal mass are obtained. From the FE model of the bare TS we also obtain its frequencies and shapes, xTS and UTS. However, the substructure of the experiment without the TS is desired. The unconstrained equations of motion to subtract the TS from the experiment are written " 2 #( ) " #( ) " T #( ) exp oexp 0 qexp f exp UFE 0 q 0 I exp , 11 TS qTS f TS q 0 I TS 0 o2 HT 0 TS TS where the exp subscript refers to the as-tested experimental system with the TS attached, and subscript TS refers to the bare transmission simulator system modeled with nite elements. To force the motions of the TS in the two uncoupled substructures to be the same, begin with the constraint
meas 0, xmeas exp xTS

12

where the constraint will be applied at only measurement DOF. The physical motions are approximated with the modal substitution yielding
meas Umeas exp qexp HTS qTS 0:

13

Now the constraint is enforced weakly by pre-multiplying with the pseudo-inverse of the TS mode shape matrix as meas y meas y meas HTS Uexp qexp Hmeas HTS qTS 0, 14 TS which is further reduced to meas y meas HTS Uexp qexp IqTS 0:

15

Eq. (15) provides the constraint to be applied to the uncoupled equations of motion, Eq. (11). When that eigenvalue problem is solved, it provides thex2 expTS and the UexpTS as inputs for Eq. (1) and Eq. (6). A constraint equivalent to Eq. (10) can be derived. It can be shown that after the transmission simulator has been subtracted from the experiment using Eq. (15), the new mode shapes are ( ) " #" # UexpTS Uexp 0 Rexp W, 16 UTSTS HTS RTS 0 where R is the coupling matrix (see Appendix A) and W are eigenvectors from the solution of the coupled subtraction problem resulting from Eq. (11) and Eq. (15). Now after the constraint is enforced, the transmission simulator and the experiment mode shapes are equal, i.e.

UexpTS UTSTS Uexp Rexp W HTS RTS W:

Substituting the last term into Eq. (10) gives h
meas Hcoin xcoin Hmeas TS TS RTS WqexpTS 0, FE HTS

17

iy

18

which can be reduced to a constraint equivalent to Eq. (10) as h iy

Hcoin xcoin TS FE RTS WqexpTS 0:

19

60

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

2.3. Diagnostics for identifying and eliminating indenite mass and stiffness matrices in the experimental substructure There is the potential for the experimental mass and stiffness matrices to become indenite. This potential arises because of the decoupling of the TS in Eq. (11). A non-positive denite matrix can be avoided in the TSM. The derivation to avoid indenite matrices begins by utilizing the coupling matrix, R, (see Appendix A). The coupling matrix, R, is used to constrain the substructures through the mass, stiffness (and damping) matrices of Eq. (11), with ( ) qexp 20 Rqc , qTS where qc represents a generalized coordinate for which the substructures are constrained to move together, (see Appendix A). Substituting Eq. (20) into the unconstrained equations of motion, Eq. (11), and pre-multiplying by RT yields " 2 # " # xexp 0 I exp 0 T T R Rqc R Rqc 0 ITS 0 x2 TS 2 3( ) 21 UT 0 f exp 5 exp : RT 4 f TS HT 0 TS Relabeling for convenience the following: " 2 # xexp 0 T R, Kc R 0 x2 TS " # I exp 0 R, M c RT 0 I TS 2 3( ) UT 0 f exp exp T4 5 fc R : f TS HT 0 TS Now, the eigenvalue problem is solved for Eq. (21) as K c o2 c M c Uc 0,
2 c 2 expTS

22

23

and where Uc are mass normalized. This eigensolution would provide the inputs for Eq. (1) and Eq. (6) with o o Uc UexpTS . However, there is still the possibility of indenite matrices. If UT c M c Uc is found to be indenite, it has been found that it can be made positive denite by throwing out some of the modes of Uc . Therefore, a reduced set of eigenvectors and corresponding eigenvalues is selected from Uc . The reduced set is labeled Uc, r . The columns removed from Uc to form Uc, r , correspond directly to the number of the diagonal entries of UT c M c Uc that have values not equal to one (typically equal to negative one). Thus, a reduced basis of eigenvectors that 2 provide a positive denite mass matrix are chosen from Uc and corresponding o2 c to be used for UexpTS and oexpTS as inputs for Eq. (1) and Eq. (6). 3. Verication Examples of the Enhanced TSM This section is intended to provide two verication examples of the new transmission simulator method. The rst example is a cylinder-plate-beam system that has been used in previous substructuring work [9,13,14]. The second example is an actual application problem. It consists of a shell FE substructure that is connected through a bolted ange to a complex internal payload experimental substructure. Both examples illustrate the effectiveness of the method. 3.1. Cylinder-plate-beam substructuring example A cylinder-plate-beam problem is used to validate the new transmission simulator method. The problem consists of a cylinder substructure, Fig. 1(a), that is to be connected to a plate, beam, and TS substructure, Fig. 1(b); however, the transmission simulator, Fig. 2, will be decoupled from the plate, beam, and TS substructure as was described in Section 2 to form a combined system with no TS, Fig. 1(c). The physical systems are joined by eight bolts, equally spaced around the circumference of the plate. The transmission simulator, Fig. 2, was designed with four tabs to add mass and to bring the bending and axial modes of the experimental system into the testable bandwidth. The transmission simulator is connected with the bolted interface as would be done for the cylinder substructure. Therefore, the joint physical characteristics are captured within the experimental model. The experimental substructure set-up is shown in Fig. 3. The TS was instrumented with 12 triaxial accelerometers, see Fig. 2. The six rigid body modes were obtained from mass properties measurements of the parts. Nineteen other elastic

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

61

z x y x

z y

z x y

301 Z

1000 Y

Fig. 1. Cylinder-Plate-Beam System and the Substructures. The two drive points for the FRF are noted (1000Y and 301Z). (a) Cylinder substructure. (b) Plate, beam, and TS substructure. (c) Assembled system.

Fig. 2. Transmission Simulator (blue dots on left show triax accelerometer locations). (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)

Fig. 3. Modal test setup for experimental substructure with TS.

modes up to 4,000 Hz were extracted from FRFs using the multi-reference SMAC algorithm [15,16], resulting in a total of 25 modes for the plate, beam, and TS system shown in Fig. 3. Several were bending modes of the beam on the plate. Figs. 4 and 5 show the comparisons of the new TSM with an analytical result derived from a nite element model of the combined system. The nite element model of the combined system was validated against test results of the combined system. There are two TSM results shown in Figs. 4 and 5. The rst is with the plate-beam-TS substructure derived from experimental

62

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

Fig. 4. Drive Point Frequency Response Function at 301Z (lateral direction).

Fig. 5. Drive Point Frequency Response Function at 1000Y (axial direction).

results and labeled Exper-TSM in the gures. The second result is with the plate-beam-TS substructure derived from a detailed nite element model and labeled FEA-TSM in the gures. The results are very promising for the method. The old method, MCFS, also demonstrated similar results. However, this new method allows for an easy implementation into a nite element code.

3.2. Shell and internal payload application substructuring example The next example consists of a shell substructure connected though a bolted ange to a complex internal payload substructure that is depicted notionally in Fig. 6(a). The shell substructure is analytically modeled in a nite element program and the internal payload is the experimental substructure. The TS is connected to the internal payload at the ange, see Fig. 6(b), that would connect to the shell ange. The procedure described in Section 2 is followed to get a full system response. Forty-two measured responses were made on the transmission simulator which were determined by minimizing the condition number of the mode shape matrix for 33 free modes of the transmission simulator analytical model. Experimental FRF truth measurements on the full system were acquired with lateral and axial driving point hammer measurements at the base of the shell (at points 101Y and 105X), see Fig. 6(a). The axial hammer impacts and responses are in the x-direction. The lateral hammer impacts and responses are noted in the y-direction. The method of removing modes to keep the mass matrix positive denite was applied in this example. Two modes were removed.

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

63

Fig. 6. Depiction of shell and internal payload systems. (a) Cutaway of full system (force inputsred, acceleration responsesblue). (b) Experimental substructure connected with a TS at the ange. (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)

Fig. 7. Various FRFs for shell and internal payload with 2% damping. The truth data is from experiment. The TSM is with an experimental model of the payload and a FE model of the shell. The FE model is with everything modeled in FE. (a) Lateral (101Y) Drive Point FRF at base. (b) Axial (105X) Drive Point FRF at base.

Fig. 7 depicts the results. In these gures, there are three curves. The blue curve labeled Truth Data is the FRF from the experiment of the full system. The red curve labeled TSM is the TSM with the payload derived experimentally and a FE model of the shell. The green curve labeled FE Model is the full nite element model of the system. Fig. 7(a) illustrates the lateral drive point FRF at the base of the shell. Fig. 7(b) is for the axial drive point FRF at the base of the shell. Due to the complexity of the internal payload, the model has not been fully validated. In Fig. 7 it is observed that the full system FE model has some discrepancies. The same FE model for the aeroshell was used for the TSM. It is believed that an improved shell FE model would improve the TSM results. Fig. 7 illustrates that the TSM is an improvement compared to the full system FE model. Apparently, the experimental derived data for the payload is more accurate than the current unvalidated FE model of the payload. 4. Transmission simulator design To date, the transmission simulators that have been utilized have proven to be fairly robust. There are a few lessons learned that may be shared. Four areas of the design are addressed in this section, including: 1. response sensor locations; 2. force input locations; 3. physical design; and 4. design of the associated modal test. The pseudo-inverse of the mode shape matrix of the transmission simulator is applied in the constraining equations. To keep this pseudo-inversion well-conditioned, the condition number of the mode shape matrix should be minimized. The authors chose many candidate points (on the order of three times the number needed) on the TS that could be easily

64

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

instrumented, i.e. no conict with the physical connection geometries. These points need to be chosen with the possibility that they may also be used as forcing locations, so co-linear forcing paths need to be available for many of the points. The number of mode shapes to be retained for the TS was decided. Then, a sub-optimal design was programmed in an algorithm that threw away one DOF and calculated the resulting condition number of the mode shape matrix. This was performed for all candidate DOF, and the DOF was removed that produced the lowest condition number. This was repeated until the desired number of DOF for the test was obtained. Then, the forcing DOF were selected with another algorithm. Since it is assumed that the kept modes of the TS span the space of the connection motion, it may also be assumed that forcing locations that excite the bare TS mode shapes well will be acceptable for the experiment. An algorithm was developed, which requires that every mode be excited at a DOF that has a mode shape value at least half the maximum mode shape value. The algorithm rst nds the DOF that excites the largest number of modes with this criterion. If there are multiple DOFs that tie, the user selects the DOF based on the simplicity of applying the force there. Then the algorithm nds the DOF that excites the most remaining modes. The process is repeated until all modes are excited to the required level. In addition a few extra DOF are chosen to provide robustness due to uncertainty. Guidelines for the physical TS design are given below: 1. The TS should be fabricated of the same material and with the same geometry as that interface at the connecting surfaces. This allows the experimental substructure to capture the stiffness and damping of the joint itself, which is a great advantage of the method. 2. The TS should be as simple as possible to model and fabricate. The FE modeler desires to put his effort into modeling the system, not the xture, so the design should be as simple as possible. Fabrication logistics are also important. 3. The TS is designed to stress the joint in the manner of the real system. This requires that the TS has signicant mass. In general, the approach has been to attempt to mimic a local portion of the structure to which the experimental substructure will be attached. Tabs may be added to increase the mass, but they should be stiff enough so that they exercise the joint well. The experimental substructure is basically extracted from a precise modal test, in which the mode shapes must be scaled very accurately; therefore, the nal design area is the modal test. It has been found that impact testing has been more successful than shaker testing. Impact testing can usually be accomplished to a higher frequency than shaker testing, which provides more modes in the experimental model, reducing the errors due to modal truncation. It has also been found that the force gage has been too massive for the TS on the structures tested to date. In one case, when a force gage and shaker were used, the frequencies were 10% lower than with impact testing for certain modes. This is because about half the mass of the force gage is on the test article side of the gage, and is not taken into account in the force measurement. For lightly damped systems, an open loop voltage mode shaker amplier may drop out signicantly at resonance, which provides an undesirable reduction in the signal/noise ratio for the force gage at resonance. The down side of impact testing is that it can excite nonlinearities more than random shaker testing which may make the modal extraction more difcult. One test provided a signicant lesson learned. Fig. 2 depicts a TS with tabs for added mass. In one test, when the TS was attached to the test article which was very stiff, the tabs were so soft that all the tabs had a mode at about the same frequency, and they did not exercise the connecting joint very well. The TS was redesigned with short tabs and a thicker ring so that the ring exercised the joint and no individual tab modes were encountered. This provided better results. 5. Conclusion In the present study, a new effective and efcient method for connecting experimental derived substructures to analytical derived substructures was proposed. The method uses a exible xture, termed the transmission simulator (TS), at the interface of the experimental substructure. Using the TS in the experimental substructure avoids difcult issues commonly found with experimental substructures: measuring connection rotational responses and moments, and measuring constraint modes or residual attachment modes with both translations and rotations at multiple connection points. The TS is then subtracted from the experimental substructure and the remaining experimental substructure is coupled to the analytical substructure with MPCs. This is termed the transmission simulator method (TSM). The TSM presented has several advantages. A computationally efcient and convenient implementation with MPCs is used in the TSM. The TSM allows the analyst to simulate the full system with different types of solutions (transient, nonlinear transient, modal, etc.) and different loads. The TSM captures the joint characteristics at the interface of the two substructures. It allows for joining two components with multiple connections points, and a reduction in sensors for multiple connection point subsystems. The advantage of the TSM was demonstrated on two example problems. In addition, recommendations for the TS design, including algorithms for forcing and sensing locations, were provided in this study. Appendix A. Original CMS theory for transmission simulator method The original method is illustrated with three substructures, a nite element (FE) substructure, an experimental (EXP) substructure, and the transmission simulator (TS). Additional substructures may be included by simply adding more rows

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

65

Fig. A1. Examples of Substructures with Transmission Simulator in blue. (a) Experimental substructure and (b) Finite element substructure. (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)

and columns to the matrices. The TS is attached to the experimental substructure during the modal test. Force and response measurements are gathered on the transmission simulator, which account for all connection DOF. If there are other response or forcing DOF of interest for the full hybrid model on the experimental structure, those measurements are also made. An analytical model of the TS is attached to the nite element model. Physically this is impossible, because the TS and FE model occupy the same place; but mathematically, it is possible. The TS is attached rigidly to the nite element model at as many coincident nodes as possible. The joint is actually contained within the experimental model. See Fig. A1 for a representation of example substructures. The experimental substructure on the left is to be connected inside the nite element substructure on the right at a ange. Notice that the transmission simulator on the nite element model occupies some of the same space as the ange and shell. The uncoupled equations of motion for the three substructures are written as 2 2 38 9 2 9 38 xFE 0 0 q I FE 0 0 > > < FE > = < q FE > = 6 7 exp 6 0 x2 0 7 qexp 6 0 7 4 0 I exp 5 q exp 4 5> > > :q > ; 0 0 2I TS : qTS ; 0 0 2x2 TS TS 2 T 38 A:1 9 UFE 0 0 f FE > > = < 6 7 UT 0 7 6 exp 4 0 5> f exp >, :f ; T TS 0 0 2H
TS

where q is a vector of generalized modal coordinates, f is a vector of forces, I is the identity matrix, x2 is the square of the natural frequency in (rad/sec) squared and U is the mass normalized mode shape matrix. Both the FE and experimental (subscript exp) substructures include the transmission simulator xture. The compliance of the joint is included in the o2 exp terms. The last row subtracts the effects of the two transmission simulators. The damping matrix may also be included, which will be diagonal values consisting of 2zo for the FE and experimental substructures and 4zo for the two transmission simulators which will be subtracted. The damping of the joint will be included in the 2zexp oexp terms, however, the damping has been removed in Eq. (A.1) for brevity. The mode shapes YTS are typically obtained from the FE model of the TS. Two constraint equations are written, starting with the physical DOFs, xmeas , where the measurements are made. Note, that these are not necessarily the actual connection points. Since the connection is ultimately made in the space of the modal degrees of freedom of the TS, measurement degrees of freedom that accurately capture the truncated set of TS modes are assumed adequate to describe the connection motion. The two constraint equations are meas meas x FE x exp 0, meas meas 0: x exp x TS A:2

Consider only the rst line in Eq. (A.2). The second line is treated analogously. Using the modal approximation the equation can be rewritten as
meas Umeas FE q FE Uexp q exp 0:

A:3

Multiply by the pseudo-inverse of the TS mode shape matrix to obtain

meas Hmeas Umeas Umeas TS FE q FE HTS exp q exp 0:

A:4

66

R.L. Mayes, M.R. Ross / Mechanical Systems and Signal Processing 31 (2012) 5666

Eq. (A.4) is the mathematical key for the TSM. The pseudo-inverse matrix casts the connection motion on the space spanned by the mode shapes of the transmission simulator. It also softens the constraint. For example, the connection in Figure A.8 has at least 150 degrees of freedom (25 bolts times six DOF). But the problem was solved with only 33 modes of the transmission simulator. So the 150 rows in the physical connection DOF are reduced to 33 rows in Eq. (A.4). The constraint will now be satised in a least squares sense, since there will still be more sensors than modes of the TS. Small errors in the measured mode shapes become less detrimental with this method. The constraints from Eq. (A.4) can be written in matrix vector form as 9 8 > = 0 < q FE > exp C q , A:5 > > 0 :q TS ; where 2 C 4 y y meas Hmeas Uexp TS meas y meas 3 5: A:6

Hmeas Umeas TS FE
0

0 I

HTS

Uexp

I , through a reduction matrix R as Now dene the independent coordinates, q 8 9 > < q FE > = exp Rq I, q > > :q ;
TS

A:7

and substitute Eq. (A.5) into Eq. (A.7) to yield I 0: CRq A:8

If R is chosen as a matrix in the null space of C, then constraint Eq. (A.8) is satised. Eq. (A.7) is substituted into Eq. (A.1) which is also pre-multiplied by R transpose to provide the coupled equations of motion as 2 2 3 2 3 oFE 0 0 I FE 0 0 6 7 o2 0 7RqI RT 6 0 I 0 7 RT 6 exp 4 0 I exp 5Rq 4 5 2 0 0 2I TS 0 0 2oTS 2 T 38 A:9 9 UFE 0 0 f FE > > = < 6 7 FT 0 7 RT 6 exp 4 0 5> f exp >, :f ; T TS 0 0 2H
TS

Setting the right hand side equal to zero, the new eigenvalue problem can be solved for the hybrid FE/experimental system modal parameters. References
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] R.R. Craig, Structural Dynamics: An Introduction to Computer Methods, John Wiley and Sons, New York, 1981. R.R. Craig, M. Bampton, Coupling of Substructures for Dynamic Analysis, AIAA J. 6 (7) (1968) 13131319. J.H. Ginsberg, Mechanical and Structural Vibrations, rst ed., John Wiley and Sons, New York, 2001. R.H. MacNeal, A Hybrid Method of Component Mode Synthesis, Comput. Struct. 1 (1971) 581601. K.C. Park, Y.H. Park, Partitioned Component Mode Synthesis via a Flexibility Apporach, AIAA J. 42 (2004) 12361245. S. Rubin, Improved Component-Mode Representation for Structural Dynamic Analysis, AIAA J. 13 (1975) 9951006. R.R. Craig, C.J. Chang, On the Use of Attachment Modes in Substructure Coupling for Dynamics Analysis, 81st Substructures, Structural Dynamics and Materials Conference, San Diego, March 2123 1977, pp. 8999. D.J. Rixen, A. Dual Craig-Bampton, Method for Dynamic Substructuring, J. Comput. Math. 168 (1-2) (2004) 383391. M.S. Allen, R.L. Mayes, E.J. Bergman, Experimental Modal Substructuring to Couple and Uncouple Substructures with Flexible Fixtures and Multipoint Connection, J. Sound Vib. 329 (2010) 48914906. D.R. Martinez, T.G. Carne, D.L. Gregory, A.K. Miller, Combined Experimental/Analytical Modeling Using Component Mode Synthesis, in: Proceedings of the AIAA/ASME/ASCE/AHS Structures, Structural Dynamics and Materials Conference, Palm Springs, CA, 1984, pp. 140152. T.G. Carne, C.R. Dohrman, Improving Experimental Frequency Response Function Matrices for Admittance Modeling, in: 24th International Modal Analysis Conference, IMAC XXIV, St. Louis, MI, 2006. E. Corus, O. Nicolas, Using Model Reduction and Data Expansion Techniques to Improve SDM, Mech. Syst. Signal Process. 20 (2006) 10671089. M.S. Allen, R.L. Mayes, Comparison of FRF and Modal Methods for Combining Experimental and Analytical Substructures, in: 25th International Modal Analysis Conference, IMAC XXV, Orlando, FL, 2007. R.L. Mayes, M.R. Ross, P.S. Hunter, Examples of Hybrid Dynamic Models Combining Experimental and Finite Element Substructures, in: 28th International Modal Analysis Conference, IMAC XXVIII, Orlando, FL, 2010. R.L. Mayes, S.E. Klenke, The SMAC Modal Parameter Extraction Package, in: 17th International Modal Analysis Conference, IMAC XVII, Kissimme, FL, 1999. R.L. Mayes, S.E. Klenke, Extending SMAC to Multiple Reference FRFs, in: 24th International Modal Analysis Conference, IMAC XXIV, St. Louis, MI, 2006.