P. Verboven , R. Valgaeren", M. Van Overmeire and P. Guillaume Vrije Universiteit Brussel Faculty of Applied Sciences Department of Mechanical Engineering Pleinlaan 2, B-1 050 Brussel, Belgium Email: pverbove@vub.ac.be ABSTRACT This paper discusses the applicability and features of modal analysis techniques applied to automotive exhaust systems. Modal parameters are obtained by means of shaker as well as engine excitation (running mode analysis). During the modal analysis experiments with shaker excitation, attention is paid to the choice of the excitation signals, the excitation points, the boundary conditions and finding the suitable parameter estimation technique. Next, the estimated modal model is used to perform a reduced sensitivity analysis of the exhaust system. The correlation between the results of the modal analysis and the running mode analysis is analyzed and found discrepancies are discussed. Keywords: exhaust system, modal analysis, sensitivity analysis and running mode analysis. 1. INTRODUCTION During the development of an automotive exhaust system, the knowledge of the dynamical behaviour is becoming very important in order to deal with resonance and fatigue problems during the design and optimization phase [1], [2]. The classical trial and error tests used to design and optimize an exhaust system, are replaced by more advanced methods, as finite elements and modal analysis techniques. These techniques characterize the exhaust system's dynamical behaviour, by use of models and parameters. Evaluating these parameters makes it easier to detect vibration and acoustical problems early in the design phase. Moreover, these models can be used to predict and simulate the effects of modifications of the structure on the dynamical behaviour. This will contribute to a faster development of well-designed exhaust systems. The aim of this research is to integrate the modal analysis technique in the design and optimization phase. BOSAL International Research and Development Center Dellestraat 20, B-3560 Lummen, Belgium In this contribution, a case study of an exhaust system, already in production, was performed to investigate the applicability of the experimental modal analysis technique to exhaust systems. The exhaust system studied has a total length of 4.5m and weights about 15 kg. The structural components are shown in figure 1. It is a full welded structure made of stainless steel. It consists of a front and rear assembly connected by a sleeve joint. The front assembly consists of the catalyst, intermediate muffler and the overaxle tube, the rear assembly of the rear muffler with the tailpipe. The connection with the engine is made by the downpipe. The exhaust system is suspended at two bracket positions, one between de catalyst and the intermediate muffler and the second mounted on the rear muffler. An elastic suspension is obtained by means of suspension rubbers. fj rubber Intermediate muHier catalyst __ _ .. braeket ("\ 'P bracket
rear muffler tailpipe downp1pe Figure 1. Structural components of the exhaust system. First a modal test with shaker excitation is performed (see Section 2). The obtained modal model will be used in Section 3 to improve the dynamical behaviour of the exhaust system by means of the sensitivity analysis. Section 4 deals with the running mode analysis with engine excitation. Finally, the correlation between the modal and running modes and the influences of the real working conditions are discussed in section 5. 987 2. MODAL TESTING OF THE EXHAUST SYSTEM 2.1. Test Setup and Measurement Points For the development of an experimental test method for the exhaust system the boundary conditions are of primary importance. The goal of the test is to determine the modal parameters of the exhaust system in its real mounting conditions. The engine was also included in the setup due to its important contribution to the overall stiffness. To suspend the engine, a stiff steel platform bolted to the ground was used. Rubber pads between ground and frame were applied for adequate isolation from environmental vibrations. The exhaust system itself was hanging in a stiff modular frame consisting of steel tubes as shown in figure 2. Original suspension rubbers were used. This made it possible to approximate the real suspension conditions of the exhaust system. Figure 2. Experimental setup. The measurement points were uniformly distributed over the exhaust system with a spacing of approximately 0.35m (see figure 3). One measurement point at the entry and exit of each volume was foreseen. No measurement points were chosen on the mufflers and the catalyst since their resonance frequencies are much higher than these observed during the test and so these volumes were treated as rigid bodies. The exhaust system's response was hence measured (acceleration in three orthogonal directions) in a total of 22 points (see figure 3). To capture the response Signals a set of 3 three axis accelerometers (with optional water-cooling for measurements during operation) were used. To get a complete real picture of the dynamic response the use of three axis accelerometers is necessary [2]. The accelerometers were mounted on top of the exhaust by means of welded nuts as shown in figure 4. The Influence of the change in stiffness caused by the nuts is negligible. Figure 3. Wire frame model of exhaust system with indicated measurement points. Figure 4. Fixation of accelerometers on exhaust system. 2.2. Excitation I ,., I To excite the exhaust system, two electrodynamic shakers were used. They were connected to the structure by means of small steel stingers to avoid lateral input forces. Because of the complex and long-drawn shape of the exhaust system two shakers were used. To determine the optimal excitation locations and the influence of the position of the shaker on the final modal parameters, several excitation points were chosen. The excitation points, used in the final setup, were located at the tail pipe (point A) in vertical direction (Z) and at the entry of the catalyst (point B) in lateral (Y) direction. Burst random signals (50 % bursting) were applied to both shakers during pre-test and test. This signal was suitable to cope with leakage problems and feasible non-linear behaviour. The input forces were measured with 2 piezo- electric force cells. 988 2.3. The FRF Measurements The LMS GADA-X software [3], a HP C100 workstation and a DIFA SCADAS II data-acquisition unit were used to perform the MIMO (Multiple Input Multiple Output) Frequency Response Function (FRF) measurements. The frequency range was chosen between 2 and 512 Hz. The H 1 FRF estimator yielded good quality measurements after 30 averages. During the pre-test the driving point check (see figure 7), time invariance of measurements and reciprocity check (see figure 5) were well satisfied. The influence on the dynamical system characteristics due to mass loading effect and accelerometer roving is negligible. 2.4. Verification of the Linearity Assumption Two shakers were preferred to maintain a good signal to noise ratio over the whole structure and to avoid appearances of non-linearities. The linear behaviour of the exhaust system was verified by checking Maxwell's reciprocity assumption and by varying the amplitude of the excitation signal. The reciprocity was checked between the excitation points in vertical (Z) direction and is presented in figure 5. The superimposed FRFs were estimated from measurement points A and B due to excitation with one shaker at B and A respectively. The observed reciprocity is not perfectly satisfied, but the differences both in frequencies and in the amplitude of resonances are small. Figure 5. Reciprocity check(- H ;------ H ). In figure 6 FRF measurements for 3 different settings of the shaker amplifier gain are shown. It turns out that the FRFs are almost independent of the amplitude of the excitation signal applied, which indicates a nearly linear behaviour of the exhaust system. For modal testing it is required to measure optimal linearized FRFs. The random excitation technique with the H, estimator is appropriate to average out the effect of small non-linear behaviour. 11:1- --- - ,.-_;::..:..;_--..: ~ - . .:_-- Figure 6. Linearity check for different amplifier gains. (-gain 5; ----- gain 7; - -- gain 9) 2.5. Methods Used for Parameter Estimation In this study several estimation methods, SDOF as well as MDOF techniques, were applied to determine the most appropriate technique for estimating the modal parameters of exhaust systems. As SDOF techniques, the Maximum Amplitude peak and mode picking method and Circle Fitting method were inquired. The MDOF methods used were the time domain (LSCE & LSFD) and the frequency domain (FDPI) methods [1], [3]. The quality of the estimated modal parameters obtained with the SDOF techniques was rather poor because some modes were strongly coupled as shown in figure 7. .. - - - ~ - Figure 7. Driving point (A) FRF with strongly coupled modes. Since the SDOF assumption was not valid, a MDOF approach was required. To compare the time and frequency MDOF methods, both MDOF analyses were performed in exactly the same way. Both methods estimate a good modal model for the exhaust system. However, some modes were better estimated by means of the time domain method. 989 2.6. Results of Modal Parameter Estimation A total of 132 FRFs was measured to identify the modes between 2 and 200 Hz. This frequency range mainly corresponds with the second order harmonic of the engine rotation speed between 1000 and 6000 r.p.m .. This second order harmonic is the most important contribution of the engine excitation because it corresponds to the ignition frequency of the 4 cylinder (in line) - 4 stroke engine. The frequency range up to 30 Hz is very important for vibrations due to vehicle riding along the road. These vibrations are transferred through the vehicle suspensions, car body and engine. Mode Freq. Hz Damp.% Description 1 4.5 2.5 XY -Rigid body 2 8.7 4.6 XZ-Rigid body 3 12.4 1.5 XY -Bending 1 4 16.4 3.3 XZ-Bending 1 5 23.0 1.6 X-Push-pull 1 6 31.3 1.5 XY -Bending 2 7 32.7 0.8 Bending 2 8 35.7 1.0 XZ-Bending 2 9 44.5 0.8 X-Torsion 2 10 49.7 1.3 X-Push-pull 2 11 61.7 0.6 XZ-Bending 3 12 64.5 0.8 XY-Bending 3 13 68.7 0.6 X-Push-pull 3 14 97.5 0.8 Hanger rear muffler 15 109.4 0.6 XZ-Bending 4 16 123.6 1.1 Intermediate hanger 17 132.5 1.0 XY -Bending 4 18 145.9 1.1 Bending 4 19 197.7 1.2 Bending 4 Table 1. Estimated Modal Parameters. The final modal parameters of the exhaust system were estimated by using the time domain method. First, the poles (damped natural frequencies and damping ratios) and modal participation factors were estimated using the Polyreference Least Squares Complex Exponential (LSCE) algorithm. In a second step the Least Squares Frequency Domain (LSFD) technique determines the modal vectors. To limit the number of modes to be estimated within one analysis to 1 0 modes, the estimation process was repeated for a frequency range of 32 Hz until the total observed frequency range between 2 and 200 Hz was covered. In table 1 the 19 most significant modes of the exhaust system are presented. It is important to notice that, for a standard model order of 32, the modes at 12.4 Hz, 44.5 Hz and 109.4 Hz were only found using the LSCE algorithm. The model order had to be increased up to 47 to find them with the FDPI algorithm. This indicates that the MDOF time domain technique is more efficient and appropriate to estimate the modal model of the exhaust system. The modal mode shapes principally exhibit a bending behaviour while push-pull and torsion mode shapes less apparent. The mode shapes shown in figures 8-10 present the three different types of vibration behaviour of the exhaust system. 990 Figure 8. Push-pull mode of catalyst and intermediate muffler (23.0 Hz). Figure 9. Bending mode in XY plane (32.7 Hz). ( I Figure 10. Torsion mode of overaxle tube between intermediate and rear muffler (44.5 Hz). 2.7. Validation of Estimated Modal Model Several validation tests reduced the total of 26 originally estimated modes to 19 acceptable physical modes. The orthogonality/consistency check was done using the MAC values. Figure 11 presents the MAC values for the 19 modes. Considering the low MAC values (<30%) between different modes at close frequencies, there is just a small similarity and so it concerns adjacent but different physical modes. The low values (< 30%) between two estimates of different modes at distant frequencies are a strong indication for a satisfying test set-up. A sufficient number of measurement points was used to obtain a good spatial resolution in the frequency range of 2-200 Hz. Figure 11. MAC values(%). The high values (> 85%) for the modal phase collinearity and mode complexity for all 19 modes, are a strong indication for normal physical modes. The overall damping can be considered as rather proportional to the mass and stiffness distribution. These validation tools were used to eliminate the overcomplex modes related to dubious or mathematical poles. From the mode participation in table 2 it is clear that all modes were well excited by at least one of the two shakers. The excitation of the exhaust system in two different directions ensures a good excitation of all modes and an equal distribution of the input energy. This is important in obtaining an accurate modal model to perform the sensitivity analysis. Mode Freq. Hz Input: Input: Total B:+Y A:+Z (inputs) 1 4.5 100.0 5.7 5.8 2 8.7 15.0 100.0 16.0 3 12.4 100.0 6.1 11.4 4 16.4 11.2 100.0 18.2 5 23.0 48.7 100.0 8.9 6 31.3 100.0 80.6 5.7 7 32.7 51.0 100.0 6.8 8 35.7 1.0 100.0 2.3 9 44.5 12.7 100.0 5.8 10 49.7 33.1 100.0 4.0 11 61.7 6.9 100.0 2.9 12 64.5 43.4 100.0 0.9 13 68.7 9.4 100.0 0.4 14 97.5 100.0 2.1 1.9 15 109.4 58.8 100.0 0.6 16 123.6 61.2 100.0 1.5 17 132.5 100.0 51.2 2.4 18 145.9 100.0 65.5 2.7 19 197.7 18.5 100.0 1.8 Total 32.9 67.1 100.0 Table 2. Mode Participation for the two shaker excitation. 3. SENSITIVITY ANALYSIS AND PREDICTION OF STRUCTURAL MODIFICATIONS To optimize the dynamical behaviour of the exhaust system, some structural modifications were defined. LMS GADA-X software was used to perform the sensitivity analysis and to predict the effects on the modal parameters caused by the defined modifications [3]. During this study some mass and stiffness modifications were applied to shift some of the resonance frequencies. The results were validated by way of experimental tests. Both types of modifications are now briefly discussed. 3.1. Mass modifications 991 Because it is easy to modify the structure using point masses, this approach is often used to solve small resonance problems during the design and optimization of automotive applications. The mass modifications are easily defined by the amount of mass and their location on the structure. A large number of modifications can be simulated in no time. The correlation between the predicted and measured modal parameters is well satisfied for both qualitative and quantitative aspects of the results. But the application of masses to exhaust systems has the important disadvantage of increasing the total mass of the exhaust system with 0.5 to a few kilograms. To avoid this considerable increase of mass, stiffness modifications were also considered. 3.2. Stiffness modifications The application of stiffness modifications introduces some difficulties in defining the exact physical value of the applied stiffness modification. The physical value (Nm) and the DOFs between which the modification is considered specify stiffness modifications. In this study the stiffness of the exhaust system was modified, by increasing the tube wall thickness of the overaxle tube between the intermediate and rear muffler. Due to the difficulties in defining the stiffness modification, the correlation between the predicted and measured modal parameters was less satisfied. The prediction results were only reliable to decide if the stiffness modifications were a good step in the direction of solving the problem but the quantitative correlation was not very good. Therefore it is necessary to determine good approximations for the applied stiffness modifications. The use of finite element models can help to solve this problem. 4. RUNNING MODE ANALYSIS 4.1. Test Setup and Measurements To know the influence of the real working conditions on the dynamical behaviour of the exhaust system a running mode analysis was performed. For this test the same exhaust system was mounted on the car. To perform the measurements, the car was placed on a rolling road. The exhaust system's response was measured during a constant acceleration of the engine between 1000 and 6000 r.p.m. and this during 60 seconds to ensure that all resonances had been well excited during the measurements. The response signals were captured at the same measurement points with the water-cooled three axis accelerometers. A water-cooled one axis accelerometer placed in point C in vertical (Z) direction (see figure 3) was used as reference point. The crosspower spectra between each response point and the reference point were obtained. 4.2. Results To obtain the running modes, the measurement data were processed to obtain Waterfall charts for each measurement point. As discussed in paragraph 2.6 and from the Waterfall charts it was obvious that the second order harmonic of the engine excitation was the most important one. The running modes were found from the second order sections of the Waterfall charts by a simple peak picking method. Since the lack of measurements of the excitation force these peaks could also be the result of high excitation amplitudes instead of resonances of the exhaust system. Therefore the first and fourth order sections were also analyzed and the same running modes as for the second order were found. This allows concluding that the running modes are mainly due to resonances of the exhaust system. Most important running modes of the exhaust system are presented in table 3. RM No. r.p.m. Hz Description 1 1214 40.5 X-Push-pull 2 2 1793 59.8 XV -Bending 3 3 2704 90.1 XZ-Bending 3 4 3035 101.2 X-Push-pull 3 5 3697 123.2 XZ-Bending 4 6 4359 145.2 Bending 4 7 5849 194.7 XY-Bending 4 Table 3. Second Order Running Modes. As can be seen in figures 12-14, all running modes exhibit the influences of the real working conditions since the push- pull behaviour appears in all modes, mostly in combination with bending of the exhaust system. This can be explained by the pulsating gas flow of exhaust gasses due to the exhaust strokes each 180 corresponding to the second order of the rotation speed of the crankshaft. The running modes also indicate the strong influence of the engine vibrations on the downpipe, which was almost rigid during the modal analysis tests without engine excitation. 992 Figure 12. Bending mode in XZ plane (2702 r.p.m.). Figure 13. Push-pull mode of catalyst and intermediate muffler (3035 r.p.m.). Figure 14. Bending mode (4359 r.p.m.) 5. CORRELATION MODAL AND RUNNING MODES To determine the correlation between the modal modes and the running modes, their resonance frequencies and mode shapes were compared. Regarded the bad agreement of the resonance frequencies in table 4, it is clear that the correlation is not well satisfied. Below 1 00 Hz, the resonance frequencies seriously shift downwards for about 4 or 5 Hz. Above 100 Hz, this frequency shift is not so strong anymore. The comparison results of the mode shapes between the modal and the running modes, using the MAC criterion, are poor and indicate a bad similarity between the modes obtained from both analyses. MM,Hz RM, Hz Diff., Hz Description 44.5 40.5 4.0 X-Torsion 2 & X-Push-pull2 64.5 59.8 4.7 XY -bending 3 I 90.1 I XZ-Bending 3 I 101.2 I X Push-pull3 132.5 123.2 9.3 XY-Bending 4 145.9 145.2 0.7 Bending 4 197.7 194.7 3.0 XZ-Bending 4 Tabel 4. Frequency comparison of Modal and Running Modes. The fact that the correlation is not well satisfied could be expected because the testing conditions were clearly different for both analyses. It's obvious that several influences appear during the operation conditions. The most important influences can be summarized as: a) Different boundary conditions due to a higher stiffness of the exhaust system's suspension under the car. A modal analysis of the cold exhaust system mounted under the car can easily determine this effect. b) The excitation of the exhaust system due to the engine vibrations and the pulsating flow of exhaust gasses. The excitation due to the engine can be simulated by a forced response analysis. Therefore an accurate characterization of the engine as excitation source is needed. c) The high temperatures of the exhaust gasses (800 oc) affect the stiffness and damping characteristics of the exhaust system. A modal analysis of the hot exhaust system under the car can be performed on the rolling road or using a heating installation to provide a constant flow of hot air through the exhaust system. This last option can also be used to investigate the influence of the gas flow with different velocities but has the disadvantage of loosing the real boundary conditions. 6. CONCLUSIONS This paper discussed the applicability of the modal analysis technique in the study of the dynamical behaviour of an exhaust system. As a result of this study, a suitable measurement setup to perform a successful experimental modal analysis of an exhaust system has been found. The MDOF time domain (LSCE followed by LSFD) method was the best available method to estimate the modal parameters. Based on the obtained modal model, a sensitivity analysis was used to optimize the dynamical behaviour of the exhaust system. Point masses are easy to apply but result in an important increase of the overall mass of the exhaust system. On the other hand, the application of stiffness modifications is hampered by the need of an accurate characterizing of these modifications. A running mode analysis of the exhaust system was also successfully obtained. The influences of the operating conditions were clearly indicated by the bad correlation between the modal and the running modes. It is very important to characterize these different influences to be able to convert the results of the modal analysis to the exhaust system under real operating conditions. 7. ACKNOWLEDGEMENTS This research was a cooperation between the Vrije Universiteit Brussel (V.U.B.) and BOSAL Int.. The measurements took place in the Research Center of BOSAL Int., Lummen, Belgium. The authors wish to express their gratitude to the BOSAL personnel for their assistance and equipment. 8. REFERENCES [1] Heylen W., Lammens S. and Sas P., Modal Analysis Theory and Testing, K.U.L., Dept. of Mechanical Eng., PMA, Leuven, 1995. [2] Piombo B., Belingardi G., Dardano R. and Pavese M., Automotive Exhaust Pipe: the Modal Analysis for Design and Testing, Proceedings of 4'h IMAC Conf., Los Angeles (USA), 1986. [3] LMS int., GADA-X User Manuals (rev 3.3), LMS Int., Leuven, Belgium, 1994. 993