A Wave-Based Substructuring Approach for Fast Modification Predictions and Industrial Vehicle Optimization

S. Donders1, R. Hadjit1, L. Hermans1, M. Brughmans1, W. Desmet2 1 LMS International, CAE Division, Interleuvenlaan 68, 3001 Leuven, Belgium 2 K.U.Leuven, Department of Mechanical Engineering, Division PMA, Celestijnenlaan 300B, 3001 Heverlee, Belgium email: stijn.donders@lms.be

Abstract
In the vehicle development process, design decisions are increasingly based on virtual prototypes (for example using Finite Element (FE) models), as time-to-market must be reduced, while more and more variants must be designed and assessed. Substructuring and Component Mode Synthesis (CMS) methods are useful to speed up the time of local design modifications. Traditional CMS methods involve coupling the system matrices of each substructure along all interface degrees of freedom (DOFs). A Wave-Based Substructuring (WBS) approach has been developed, that aims to find a set of basis functions that describe the dynamic behavior of the coupling interfaces. The interface displacements in the assembled system are then written as a linear combination of these basis functions. As the number of basis functions is typically much lower than the number of interface DOFs, the procedure reduces the number of variables in the matrix equations and the size of the interface description. This greatly facilitates the model reduction procedure and results in faster structural and vibro-acoustic predictions. The WBS approach has been implemented in an integrated multi-attribute virtual simulation environment, where it complements the assembly definition and trimming functionality as it allows speeding up modification predictions and thus alleviates the computational burden of industrial vehicle optimization. This is demonstrated on the basis of three test cases taken from automotive industry.

Introduction

Fast assembly predictions become ever more important in the vehicle development process. Design decisions are more and more based on virtual prototypes, as time-to-market and development costs must be reduced. The additional trend of mass customization forces engineers to design a higher number of variants on a lower number of platforms. The Finite Element (FE) method is widely used to predict the vibro-acoustic vehicle performance. FE models can only be applied in the low and medium frequency range, due to the model sizes and subsequent computational costs that grow with frequency. To partially overcome this practical limit, extensive work has been performed on substructuring and Component Mode Synthesis (CMS) techniques [1][2][3][4]. The degrees of freedom (DOFs) of each substructure are expressed in terms of a limited number of component modes; the component models are then synthesized. Recently, Automated Multilevel Substructuring (AMLS) [5][6] has been developed. A vehicle body is recursively divided into dozens of levels of in total thousands of substructures, based on the mathematical structure of the FE models, rather than on the physical composition of the system. Each substructure is separately solved and the results are synthesized. When coupling multiple (levels of) reduced systems, the interface problem size becomes increasingly dominant. To further speed up the synthesis, the interface representation size between components must be reduced. This can be done by the condensation of the interface displacements as a linear combination of a limited set of interface basis functions (waves). As the required number of basis functions is typically much lower than the number of interface DOFs, faster assembly predictions are obtained. Previous papers have
1901

1902

P ROCEEDINGS OF ISMA2006

reported on the use of component modes to derive the interface basis functions [7][8]. When one couples a very stiff component with a very flexible component, an obvious choice is to derive the interface basis functions from the component modes of the stiff component [9]. For components with similar modal densities the component modes are typically less suitable. In the Wave-Based Substructuring (WBS) [11] approach applied in this paper, one performs a single computation of the full assembly model to obtain the interface basis functions. This allows to accurately capture the interface dynamics, and with modification analysis and optimization in mind, a single full computation is not a large burden: an optimization may consist of numerous iterations involving hundreds of FE runs. In Wave-Based Substructuring, the single full run enables that the interface representation size is reduced, which facilitates the model reduction procedure of components and speeds up the assembly analysis, while the accuracy is maintained. Section 2 discusses the mathematical framework of the WBS approach, for the rigid and elastic coupling case. In the following sections, the WBS approach is demonstrated on three cases taken from automotive industry. Section 3 considers two structural analysis cases: the re-design of a vehicles B-pillars, and the modification analysis of coupling stiffness of a windscreen in a vehicle body-in-white (BIW). Section 4 then considers a vibro-acoustic case: the modification of dynamic coupling stiffness of a windscreen in a vehicle trimmed body. The paper is concluded in Section 5.

2
2.1

Theory
WBS Assembly Procedure

Consider an FE substructure in an assembled system. The DOFs x can be divided into interior DOFs xi and junction (coupling) DOFs xj. The system matrices (for the undamped case) are then partitioned into submatrices:

M ii M ji

&i K ii M ij & x + &j M jj & x K ji

K ij x i f i = K jj x j f j

(1)

In this formulation, all junction nodes have 6 DOFs. The Wave-Based Substructuring approach restricts the junction DOFs xj, by expressing them as a linear combination of a set of N interface basis functions V, weighted with participation factors p:

xj =V p
By substituting Equation (2) in Equation (1), one obtains

(2)

M ii V T M ji

M ij V & &i K ii x + T T V M jj V & & p V K ji

K ij V x i f i = T V T K jj V p V f j

(3)

When the number of basis functions N is less than the number of junction DOFs xj, this substitution reduces the size of the FE matrix equations compared to classical CMS techniques. One must then translate the equilibrium and continuity conditions in terms of the interface basis functions. As described in [11], for a rigid connection between two substructures a and b, the following equilibrium and continuity conditions apply:
b xa and j = xj

f ja + f jb = 0

(4)

In a WBS framework, similar conditions apply on the participation factors p:

p a = pb

and

f ja + f jb = 0

(5)

M ADUSE

1903

An elastic connection between substructures a and b is applied with the following equilibrium and continuity condition (where the matrices K denote the coupling stiffness matrices between the junction DOFs of the substructures):

f ja K aa b = f j K ba

K ab x a j b K bb x j

(6)

Following Equation (2), for substructures a and b, the junction DOFs are expressed as a linear combination of sets of interface basis functions Va and Vb, yielding

VaT f ja VaT K aaVa VaT K abVb p a T b = T b T Vb f j Vb KbaVa Vb K bbVb p

(7)

As the required number of basis functions is typically much lower than the number of interface DOFs, the WBS framework presented in this section results in a reduced-sized assembly definition, which allows faster assembly calculations.

2.2

Model Reduction Procedure

In general, a substructuring method contains a reduction step for at least one of the components. For example, in a vehicle optimization case which aims to reduce the acoustic radiation of a floor panel, one will be interested to apply a model reduction step on the remainder of the vehicle. This reduced-sized model of the remainder of the vehicle is then re-assembled with the FE representation of the floor panel, which enables much faster modifications and optimization on this floor panel. Such a reduction step involves reduction of the number of degrees of freedom of the substructure model, by representing the physical degrees of freedom of each substructure into a reduced number of so-called generalized coordinates. Numerous papers have reported on the selection of these generalized coordinates. All of them make use of the vibration normal modes of the substructures; they differ in the boundary conditions that are applied to the substructure, and in the selection of enrichment vectors to these normal modes. Two well-established examples are: the Craig-Bampton fixed interface approach [1], where the normal modes are computed while the substructure is clamped at the connection interface, and where the enrichment vectors consist of constraint modes (which are the static deformation shapes of the substructure obtained by successively applying a unit displacement on one interface degree of freedom, while holding the remaining interface DOFs fixed, and repeating this for all interface DOFs). MacNeals approach [10] uses the normal modes of the component in free-free conditions, and enrichment vectors that consist of residual flexibility modes (which are the static deformation shapes obtained by successively applying a unit force on one of the interface degrees of freedom, with a zero force on the remaining interface DOFs, and repeating this for all interface DOFs).

In such reduction methods, the component modes are generally computed up to a frequency higher than the frequency range of interest for the assembly (a rule-of-thumb is to increase the frequency range for the component reduction with a factor 1.5 w.r.t. the frequency range of interest). The enrichment vectors are of vital importance to accurately represent the local flexibility at the connection interface, when the reduced modal model of the substructure is re-assembled to other substructures. In the above example methods, it becomes clear that the number of enrichment vectors that must be computed, strongly (linearly) depends on the number of interface degrees of freedom. This also holds for alternative formulations, see e.g. the overview in [2]. Since the WBS approach (strongly) reduces the number of interface degrees of freedom in the assembly definition, a reduction procedure based on WBS will be much more efficient, since a much lower number of enrichment vectors need to be computed. A quantitative example of this benefit in the reduction procedure will be given in Section 3.1.

1904

P ROCEEDINGS OF ISMA2006

Structural Analysis

As a general procedure, when building the FE model of a vehicle body, the panels are meshed from CAD data and assembled together using dedicated elements to represent the connections. Numerous spot welds are applied between the panels, predominantly in the BIW assembly process. Other connection methods in the vehicle assembly process involve the creation of seam welds, sealing connections, glue, etc. LMS Virtual.Lab is an integrated multi-attribute virtual assembly and simulation environment [14], in which all of these connections between components can be applied. The WBS approach is used in this environment to isolate certain FE components of interest, to create a reduced modal model of the remainder of the vehicle (with much lower computational workload than in a conventional reduction), and then create a reduced-sized assembly model to quickly evaluate the effect of local design iterations (on the isolated component) on global performance targets (in terms of eigenfrequency, noise level, ...). This section considers two structural analysis cases: the re-design of a vehicles B-pillars (involving WBS with rigid connection) and the modification analysis of coupling stiffness of a windscreen in a vehicle BIW (involving WBS with elastic connection). Section 4 then considers a vibro-acoustic application.

3.1

B-Pillar Re-design

In the first industrial application case, Wave-Based Substructuring is applied to speed up the re-design of the B-pillars in an industrial BIW model, for which the FE model (274.338 nodes, 209.328 elements) is shown in Figure 1 (left). The aim is to use WBS to re-design the dynamic characteristics of the b-pillars in the range 0-100Hz, so that a reduced modal model must be computed for the remainder of the vehicle (as shown in Figure 1 (right). In this industrial-sized example case, there are 1788 physical connection DOFs, a number of which have been defined at the spot weld level. Note that the WBS approach has been adapted for the substructuring of the FE body model (BIW or Trimmed Body) at the spot weld level [12][14]. The substructures are defined from a set of panels and the degrees of freedom of the spot weld interface are replaced by a set of basis functions - the waves.

Figure 1: B-pillar re-design case: Body-in-White FE model (left), and WBS Reduced assembly model (right), in which the B-pillars are kept in FE representation, and a reduced modal model has been created for the remainder of the vehicle body.

M ADUSE

1905

In the frequency range up till 100Hz, only 35 wave degrees of freedom are required in the WBS assembly connection, whereas a conventional assembly would consider 1788 physical connection DOFs. For the full FE model, the natural modes calculation up till 100Hz requires 1h42min of real CPU time. For the modal reduction step for the remainder of the vehicle body, o The WBS reduction procedure an adaptation of MacNeals approach [10] requires 2h28min to compute the natural component modes up till 150 Hz, as well as the enrichment vectors for the 35 wave degrees of freedom. As a comparison, the reduction procedure (using the original approach of MacNeal [10]) has also been attempted that is, to calculate the natural component modes up to 150 Hz as well as the enrichment vectors for all of the 1788 physical degrees of freedom. This calculation was not feasible on the workstation available for the job. The calculation of (only) half the required number of enrichment vectors was feasible, but required already much more time (more than 7 hours) just to get halfway the set of enrichment vectors! Moreover, this calculation time increases more than linearly when the number of enrichment vectors is increased.

The WBS assembly structure (full FE b-pillars with reduced modal model for the remainder of the vehicle, as shown in Figure 1, right) calculates in 60s, so 100x faster than the full natural modes calculation for this model. When comparing the results, it is clear that the WBS assembly structure is very accurate: o o the relative eigenfrequency difference is less than 1.2 10-4 for all modes up till 100Hz the MAC [15] is larger than 0.9998 for all modes up till 100Hz, see also Figure 2.

The above quantification clearly illustrates that the WBS approach is beneficial for the reduction procedure. It obviously increases the efficiency and feasibility of this vital step. It thus enables the creation of the reduced-sized assembly model (the WBS assembly structure in Figure 1, right), which evaluates much faster than the full FE analysis model, with the same accuracy in the results.

Figure 2: MAC to compare the modes of the full BIW (Figure 1, left) with the modes of the WBSreduced model (Figure 1, right). Clearly the results are very accurate.

1906

P ROCEEDINGS OF ISMA2006

Next, it will be verified whether one can obtain accurate modification predictions on the basis of this reduced-sized WBS assembly structure. As modification, the number of spot welds along the length of the b-pillars is drastically modified. The nominal design has been created in LMS Virtual.Lab [14], with 4 line connections with re-distributed spot weld properties (at an average spot weld distance). For the nominal design, 17 spot welds have been created along each line, as shown in Figure 3 (left). For the modified design, the number of spot welds along each line is drastically reduced to only 6 or 7 per connection line.

Figure 3: B-pillar re-design: change the number of Hexa spot welds along 4 connection lines along the length of the b-pillars, using LMS Virtual.Lab [14]. In the nominal design (left), the average spot weld distance is 55mm, so that 17 spot welds are created along each line; in the modified design (right), the distance is increased to 150mm, so that only 6 to 7 spot welds are created along each line. For this modification, Figure 4 compares the full FE analysis results for the nominal model, with the full FE analysis results for the model, in which the B-pillars have been modified as in Figure 3, left (so with 6 or 7 spot welds, instead of 17 spot welds, along each line). Clearly the dynamics have changed as a result of the modification. To assess the quality of the modification prediction with WBS, a comparison is made between the modes of the BIW with modified B-pillars (as in Figure 3, right), obtained from a full FE validation analysis of the modified model, and from a fast WBS-reduced analysis: o o o the relative eigenfrequency difference is less than 0.3 %for all modes up till 100Hz the MAC [15] is larger than 0.98 for all modes up till 100Hz, see also Figure 5. Figure 7 demonstrates the accuracy of WBS when used for a structural FRF calculation (frequency range up till 100Hz, with input at the engine mount, DOF +Z, and output at the driver seat rail, DOF +Z). The nominal FRF (for the model with 17 spot welds along each line (blue, dotted curve) is compared with the FRFs obtained with the modified models (with 6 or 7 spot welds along each line): the prediction with the WBS-reduced model (black, solid) and with a full FE modification analysis (red, dashed). It can be seen that the effects of the modification are substantial (the nominal curve clearly differs from the modified curves), and that the WBS-reduced modification prediction is very accurate when compared to the full FE modification prediction, for both amplitude and phase.

From the above comparison, it can be seen that the modification prediction with the WBS-reduced assembly structure is very accurate. So despite drastic changes in the B-pillar design, WBS can be used to predict the modified BIW dynamics in a very accurate and efficient manner (recall that for this case, the WBS prediction is 100x faster than a full FE prediction of the B-pillar modification).

M ADUSE

1907

Figure 4: B-pillar re-design: MAC to compare the nominal modes (B-pillars as in Figure 3, left) with the modes obtained when the B-pillars are modified (Figure 3, right). Clearly the dynamics have changed as a result of the modification.

Figure 5: B-pillar re-design: MAC to compare modes of the BIW with modified B-pillars (as in Figure 3, right), obtained from a full FE analysis and from a WBS-reduced analysis. Clearly the WBS results are very accurate.

Figure 6: B-pillar re-design: FRF from engine mount (+Z) to seat rail (+Z) in the range 0-100Hz. For the nominal model (with 17 spot welds along each line), the blue, dotted curve is obtained. For the modified models (with 6 or 7 spot welds along each line), the prediction with the WBS-reduced model (black, solid) is compared with the full FE modification prediction (red, dashed). The modification is substantial (the modified curves are different from the nominal curve), and the modification is accurately predicted with WBS (the modified curves clearly overlap) Of course, a variety of other modifications on the b-pillar can be considered, such as modifications of shell thickness, material properties (Youngs modulus, mass density), but also re-inforcements to locally strengthen the B-pillar, shape modification (cross section, morphing) and even shape optimization. Note that elsewhere in these proceedings, you can find the paper [16] which deals with the same b-pillar application model. This paper studies the effect of uncertain input parameters (spot welds, reinforcements, ...) on the static performance of the industrial vehicle model, by combining the fuzzy FE method with static substructuring based on Guyan reduction [17].

1908

P ROCEEDINGS OF ISMA2006

3.2

Windscreen-BIW Coupling

Figure 7 shows the second industrial application case: the coupling of a windscreen in a vehicle body-inwhite (BIW). The connection between the windscreen and the vehicle BIW is realized with glue. The glue has been modeled using elastic springs defined between 149 pairs of nodes and for the translational DOFs only. A modification analysis is performed, multiplying the translational coupling stiffness with a factor ranging from 0.1 to 10. Such modifications have a substantial effect on the vehicle body dynamics. Figure 8 shows a MAC [15] comparison (top view) in LMS Virtual.Lab [14] between the nominal BIW+windscreen assembly modes and the modes obtained when the elastic coupling stiffness is multiplied with 10. The dynamics have changed significantly: the MAC clearly deviates from unity for a large number of modes.

Figure 7: BIW-Windscreen Coupling

Figure 8: MAC to compare the nominal BIW+ windscreen modes with the modes obtained when the elastic coupling stiffness is multiplied with 10. Clearly the dynamics have changed as a result of the modification.

It is assessed whether Wave-Based Substructuring can be used to obtain fast and accurate predictions in the above-mentioned modification range. 92 interface basis functions are obtained from the normal modes of the nominal BIW+windscreen assembly. As the number of basis functions is less than the number of junction DOFs (447), the WBS approach reduces the problem size. Figure 9 and Figure 10 compare the conventional assembly modes with the WBS predictions up to 150 Hz. Figure 9 shows that with a factor 0.1, the MAC diagonal is larger than 0.8 for all modes, and the frequency difference remains below 0.5%. Likewise, Figure 10 shows that with a factor 10, the MAC diagonal is larger than 0.85 for all modes, while the frequency difference never exceeds 0.3%. So despite drastic changes in coupling stiffness, it can be concluded that WBS can be used to accurately and quickly predict the BIW+windscreen assembly dynamics. This example shows that a WBS assembly of reduced modal models can be used to efficiently perform sensitivity analyses to connector properties modifications. It gives very accurate results and, as reduced component models are used, the analyses are very fast.

M ADUSE

1909

Figure 9: BIW-Windscreen case, translational stiffness modification with a factor 0.1: WBS vs. full FE results, in terms of frequency (top), relative eigenfrequency difference (middle) and MAC diagonal (bottom).

Figure 10: BIW-Windscreen case, translational stiffness modification with a factor 10: WBS vs. full FE results, in terms of frequency (top), relative eigenfrequency difference (middle) and MAC diagonal (bottom).

Vibro-Acoustic Analysis

In this section, Wave Based Substructuring has been applied to a vibro-acoustic analysis case, with the aim to reduce the interface description of a windshield to trimmed body connection. This coupling has dynamic stiffness as shown in Figure 11 (defined in the range [10,150]Hz, and varying between 0.2 and 5 times the nominal stiffness value). Consider the acoustic cavity in Figure 12. The aim is to predict the Noise Transfer Function (NTF) between the structural input point (engine head mount) and a target pressure output point (front center microphone).

Figure 11: Frequency dependent spring properties of trimmed bodywindscreen connection. For the structural part, two assembly models are created:

Figure 12: Problem outline: noise transfer function from structural input point to output microphone.

Conventional assembly of full FE body and windscreen. Acoustic Transfer Vectors (ATV) are used for the vibro-acoustic calculation [18]; WBS assembly of reduced modal models of body and windscreen. Modal Acoustic Transfer Vectors (MATVs) are used for the vibro-acoustic calculation [18].

1910

P ROCEEDINGS OF ISMA2006

For a given frequency, direct forced response analysis [13] is used to compute structural responses (for the full model) or modal responses (for the WBS assembly). The structural responses are then projected onto the acoustic mesh, and LMS Virtual.Lab [14] Acoustics can then be used to predict the microphone sound pressures p based on the LMS SYSNOISE [19] solver, using Equation (8a) for the full model and using Equation (8b) for the WBS assembly, so that the results can be compared.

p = {ATV }{v}

(8a) (8b)

p = {MATV }tb {MRSP}tb + {MATV }ws {MRSP}ws

with v the structural normal velocity and MRSP the structural modal responses (vector of the modal participation factors). Subscripts tb and ws denote trimmed body and windscreen, respectively. On an SGI Onyx3 workstation (1600 MHz MIPS R14000, 2GB memory), the structural solution for the conventional assembly requires 423s, and for the reduced WBS assembly it is obtained in 88s. The model considered here has a coarse mesh; for a refined mesh, the gain in time would be even more important. The calculation time of the full model rapidly increases with the model size, while it stays approximately in the same range for the reduced assembly. OPTIMUS [20] is a commercial software package for process integration and design optimization. Using the process integration functionality, the above-mentioned computation process has been captured with the frequency as a variable. This allows to easily generate comparison results for discrete frequencies. Figure 13 compares the results for the full FE assembly structure with the results of the reduced WBS assembly structure, for the range of discrete frequencies 10, 11, 12, 150Hz. It can be seen that the results are identical, as the curves are superimposed. It can be concluded that the reduced WBS assembly can be used to accurately and efficiently predict the full FE assembly results. To underline the industrial relevance to accurately model the dynamic stiffness, the acoustic pressure has also been predicted at the same discrete frequency lines with a constant nominal coupling stiffness (in Figure 11, this corresponds to a factor 1 in the entire range). Figure 13 compares the constant stiffness curve (for the full FE model) with the dynamic stiffness curves (for the full FE model and the WBS reduced model). Clearly the effect of frequency dependent coupling properties is important to accurately predict the acoustic pressure in the range above 60 Hz.

Figure 13: Coupling of windscreen in trimmed body: pressure response obtained with dynamic stiffness for the full FE assembly (red, dashed) and the WBS assembly (thin black, solid) the curves clearly overlap. For comparison, the pressure response obtained with constant stiffness is also shown for the full FE assembly (blue, dotted) clearly a different curve.

M ADUSE

1911

Conclusions

A Wave-Based Substructuring (WBS) approach has been developed to enable fast vehicle body optimization. An efficient assembly formulation is obtained by writing the interface displacements as a linear combination of interface basis functions. As the required number of basis functions is typically much lower than the number of interface DOFs, faster assembly predictions are obtained. Interface basis functions are obtained from a single normal modes solution of the full assembly model. These basis functions accurately capture the dynamics of the interface in the assembly, so that the interface DOFs can be expressed accurately in terms of these basis functions. Performing a single full analysis may sound counter-intuitive, as substructuring and condensation methods were originally developed to prevent having to perform a complete assembly analysis. However, in a modification analysis and optimization framework, one might have to perform hundreds of iterations. It is then most critical to speed up the time required for a single iteration, so that a single full computation that allows doing so is not a large computational burden. The WBS approach has been worked out for the rigid and elastic coupling case, and can be applied to efficiently and accurately assemble FE components and/or reduced modal models of components. It has been shown that the WBS predictions are very accurate when compared to the full FE predictions. Three industrial scenarios have been considered, involving both structural analysis and vibro-acoustic performance: the re-design of the B-pillars, the coupling of a windscreen in a vehicle body in white using linear elastic stiffness, and the coupling of a windscreen in a vehicle trimmed body model using dynamic stiffness properties. Results from the different analyses showed that the WBS approach yields very efficient and accurate predictions of elastic coupling stiffness modifications, even when drastic modifications are applied. Comparison of the computation time showed the benefits of combining modal reduction and Wave-Based Substructuring. A key benefit of the WBS approach is that it by reducing the number of interface degrees of freedom greatly facilitates the reduction procedure for components (as in such a reduction procedure, typically one must compute enrichment vectors one for each interface degree of freedom between components). This benefit has been quantified on the basis of an industrialsized example case. For these industrial cases, it has been shown that the WBS results compare well with full FE predictions in a wide modification range in terms of accuracy and CPU time. This proves the validity of WBS for performing accurate and efficient optimization within the considered modification range.

Acknowledgements
The work presented in this paper has been performed in the framework of the ongoing research project Analysis Leads Design Frontloading Digital Functional Performance Engineering, which is supported by IWT Vlaanderen. In addition, MMC is acknowledged for their participation to the research and development.

References
[1] [2] [3] R.R. Craig Jr., Structural dynamics an introduction to computer methods, Wiley, NY, (1981). D.-M. Tran, Component mode synthesis methods using interface modes. Application to structures with cyclic symmetry, Comput Struct, Vol. 79 (2001), pp. 209-222. M.A. Tournour, N. Atalla, O. Chiello and F. Sgard, Validation, performance, convergence and application of free interface component mode synthesis, Comput Struct, Vol. 79 (2001), pp. 1861-1876.

1912

P ROCEEDINGS OF ISMA2006

[4] [5] [6]

L. Hermans and M. Brughmans, Enabling vibro-acoustic optimization in a superelement environment: a case study, Proc. IMAC XVIII, San Antonio, TX, USA, (2000), pp. 1146-1152. J. Bennighof, M. Kaplan and M. Muller, Extending the frequency response capabilities of automated multi-level substructuring, AIAA-2000-1574, (2000). A. Kropp and D. Heiserer, Efficient broadband vibro-acoustic analysis of passenger car bodies using an FE-based component mode synthesis approach, J Comput Acoust, Vol. 11, (2003), pp. 139-158. G. Sandberg, A new strategy for solving fluid-structure problems, Int. J. Numer Meth Eng, Vol. 38 (1995), pp. 357-370. E. Balms, Use of generalized interface degrees of freedom in component mode synthesis, Proc. IMAC XIV, Dearborn, Michigan, USA (1996), pp. 204-210. L. Ji, B. Mace and R.J. Pinnington, A hybrid mode Fourier-transform approach for estimating the vibrations of beam-stiffened plate systems, J Sound Vib, Vol. 274, (2004), pp. 547-565.

[7] [8] [9]

[10] R.H. MacNeal, A hybrid method of component mode synthesis, Comput Struct 1 (1971), pp. 581-601. [11] S. Donders, R. Hadjit, K. Cuppens, M. Brughmans and W. Desmet, A Wave-Based Substructuring Approach for Faster Vehicle Assembly Predictions, Proc. NOVEM 2005, SaintRaphal, France, April 18-21, (2005). [12] R. Hadjit, M. Brughmans, H. Shiozaki, Application of Fast Body Optimization Procedures to Shorten Car Development Cycles, Proc. JSAE 2005, Yokohama, Japan, May 18-20, (2005). [13] MSC, MSC.Nastran 2004, (2004). [14] LMS International, LMS Virtual.Lab Rev 6A, (2006). [15] W. Heylen, S. Lammens and P. Sas, Modal Analysis Theory and Testing, K.U. Leuven, Department of Mechanical Engineering, Division PMA, Leuven, Belgium, (1997). [16] L. Farkas, D. Moens, S. Donders, D. Vandepitte, W. Desmet, The fuzzy FE approach to assess the uncertain static response of an industrial vehicle, Proceedings of ISMA 2006, Leuven, Belgium, September 18-20, (2006). [17] R. Guyan, Reduction of stiffness and mass matrices, AIAA Journal Vol. 3, No. 2, (1965), pp. 380-387. [18] F. Gerard, M. Tournour, N. El Masri, L. Cremers, M. Felice, A. Selmane, Numerical Modeling of Engine Noise Radiation through the use of Acoustic Transfer Vectors - A Case Study, Proc. SAE 2001-01-1514, Traverse City, MI, USA, (2001). [19] LMS International, LMS Sysnoise Rev 5.6, (2003). [20] Noesis Solutions, OPTIMUS Rev. 5.0, (2004).