Optimization process for correlation of experimental and numerical modal analysis

Transcription

1 Optimization process for correlation of experimental and numerical modal analysis European HyperWorks Technology Conference EHTC 2011 November 8 th 9 th, 2011, Bonn, Germany Dr.-Ing. Martin H. Müller-Bechtel, Dipl.-Ing. Simon Tschirpke, TECOSIM Technische Simulation GmbH

2 Forecast Forecast TECOSIM TECOSIM: Facts & Figures Foundation: 1992 Business Area: Markets: Vision: TECOSIM is Europe s largest, independent service provider in Computer Aided Engineering (CAE) the computer-based development and optimization of components, structures and products Original Equipment Manufacturers (OEMs) and suppliers in following industries - automotive - aerospace - chemical - energy - railway - medical Global Leader in Computer Aided Engineering Turn over (Mio. ) ,8 7,7 8, Employees

3 Aim: verification of NVH simulation models Essential for virtual vehicle development Complex dependencies to wide range of parameters Method: Usage of optimization for parameter identification Better correlation in less time Correlation should consider frequency and mode shape: Relative frequency of simulation mode compared to test MAC (modal assurance criterion) value for comparison of mode shapes 2 a i bj MAC with a i shape vector of test mode i i,j a a b b b j shape vector of sim mode j i i j j 4/22

4 Typical NVH models: coupled structures Full vehicle: modes of suspension, engine Trimmed body: coupled masses like cooler, battery, or mass damper Body in white: with elastically connected subframes Full vehicle mode Subframe mode ML BiW with subframe mode 2 5/22

5 Typical parameters to identify: Coupling element stiffness (bushings, engine mount, ) Joining technique: modeling method dependent properties: glue material stiffness, weld line thickness Properties for not exactly known materials like: fiber reinforced plastics Distribution of masses of hang on parts Structural glue Laser weld Weld line 6/22

6 Example model: rear subframe out of TEC BENCH process Purchasing rear subframe Hardware NVH testing Scanning FE simulation FE modeling Geometry data preparation 7/22

7 Example model: rear subframe out of TEC BENCH process Initial correlation status: Test frequency correlation status MAC value rel. Frequency 97% 103% 1% 141% 86% sim Nearly perfect match for 1 st and 2 nd mode (MAC > 0.99) Reduced frequency for 5 th mode (mass damper effect) Parameters to identify: Bushing stiffness Distribution of masses inside rubber bushings 8/22

8 Example model: rear subframe out of TEC BENCH process Criterion for correlation considers frequency AND mode shape: MAC value for comparison of mode shapes Relative frequency of simulation mode compared to test F i,j penalty function with gradient to correlation window ±5% mid section uses Runge function Correlation criterion: F i,j MAC i,j frequency filtered MAC value Test frequency correlation status MAC value rel. Frequency 97% 103% 61% 141% 86% sim /22

9 Example model: rear subframe out of TEC BENCH process Correlation criterion considers best matching simulation mode: For each test mode take the best matching simulation mode: maxfmac i = max(f i,j MAC i,j ) Optimization objective: Maximize correlation of best matching simulation modes for all test modes maximize max fmac i Test frequency correlation status MAC value rel. Frequency 97% 103% 61% 141% 86% max sim Test frequency correlation status MAC value Σ rel. Frequency 97% 103% 61% 141% 86% sim /22

10 Setup of optimization process in HyperWorks Method used: TCL programming Implementation of test data: Reading test data file in csv format Measurement point IDs (9) Node IDs Nodes SET Test modes (5): Frequency DTABLE entry (5) Test modes (5): shape vectors (27) DTABLE entries(135) Test Frequency amplitude Shapes E E E E E E E E-02 11/22

11 Setup of optimization process in HyperWorks Method used: TCL programming Response definition for simulation modes: EIGRL card with fixed number of modes (15) Simulation modes (15): Frequency DRESP1 entry (15) Simulation modes (15): shape vectors (27) DRESP1 entries (405) Δ Mode specific displacement response not supported by OptiStruct! Solver switched to NASTRAN Δ Mode specific displacement response not defined in NASTRAN template! Direct programming of NASTRAN code (replacement.nas) referencing replacement.nas in bulk unsupported section moving of DRESP1 to separate include (waste_basket.nas) 12/22

12 Setup of optimization process in HyperWorks Method used: TCL programming Calculation of single frequency filtered MAC: Definition of DEQATN Parameters out of DTABLE: Test shape (27), test freq (1) Parameters out of DRESP1: sim shape (27), sim freq (1) Equation for MAC calculation Frequency filtered MAC test (5) vs. simulation (15): F i,j MAC i,j DRESP2 (75) DEQATN 2 f( + t001x,t001y,t001z, + + t009x,t009y,t009z, + T, + s001x,s001y,s001z, + + s009x,s009y,s009z, + S)=(( + t001x*s001x+t001y*s001y+t001z*s001z+ + + t009x*s009x+t009y*s009y+t009z*s009z + )**2/(( + t001x*t001x+t001y*t001y+t001z*t001z+ + + t009x*t009x+t009y*t009y+t009z*t009z + )*( + s001x*s001x+s001y*s001y+s001z*s001z+ + + s009x*s009x+s009y*s009y+s009z*s009z + ))) + *MIN((S/(0.95*T))**4,(T/(0.95*S))**4, /(1.+25.*(S/T-1.)*(S/T-1.))) 13/22

13 Setup of optimization process in HyperWorks Method used: TCL programming Selection of best correlating sim mode for each test mode: Definition of DEQATN Parameters out of DRESP2: test mode related simulation mode filtered MACs (15) Selection of maximum MAC for each test mode DRESP2 (5) ΔDRESP2 referencing DRESP2 not supported by NASTRAN template NASTRAN template modified with specific section out of OptiStruct template DEQATN 4 f( + m1, + + m15, + )= + Max( + m1, + + m15, + ) 14/22

14 Setup of optimization process in HyperWorks Method used: TCL programming Sum of best correlating frequency filtered MAC values: Definition of DEQATN Parameters out of DRESP2: maximum MAC for each test mode (5) sum of maximum MAC for each test mode DRESP2 (1)! DRESP2 referencing DRESP2 now supported by modified NASTRAN template Definition of design objective DEQATN 3 f( + m1, + + m5 + )= + m m5 Sum of maximum MAC DESOBJ(MAX) (1) 15/22

15 Setup of optimization process in HyperWorks Method used: manual definition Request for optimization output: Output of responses restricted to frequency and best correlating filtered MAC DSAPRT (1) Δ DSAPRT not supported by HyperWorks Manual definition in unsupported section Definition of responses for output SET (1) Δ SET of responses not supported in NASTRAN template Manual definition in unsupported section 16/22

16 Setup of optimization process in HyperWorks Method used: manual definition Problem specific definition of design variables: Stiffness of front/rear bushing 6DOF DESVAR (12) Stiffness of rubber bands 3DOF DESVAR (3) PBUSH-property relations DVPREL (15) Δ DVPREL for PBUSH not supported by NASTRAN Template Direct definition of NASTRAN cards (replacement.nas) moving of DVPREL to separate include (waste_basket.nas) Mass distribution for bushing bolts DESVAR (4) PMASS-propery relations DVPREL (4) Δ DVPREL for PMASS not supported by NASTRAN Template Direct definition of NASTRAN cards (replacement.nas) moving of DVPREL to separate include (waste_basket.nas) 17/22

17 Example model: rear subframe out of TEC BENCH process Optimized correlation status: Test frequency correlation status MAC value rel. Frequency 97% 98% 103% 100% 126% 1% 121% 141% 86% 98% sim Nearly perfect match for 1 st and 2 nd mode (MAC ~ 0.99) Improved match of frequency and shape for 5 th mode (mass damper effect) Parameters identified: Bushing stiffness Distribution of masses inside rubber bushings 18/22

18 How does HyperWorks support the process? All keywords necessary for the process are supported by HyperWorks Tcl programming interface allows for very efficient setup of a huge number of entities. Tcl programming benefits of command logging (used like macro recording for easy access to command syntax). Different ways of work around But there is a variety of entities not or not fully supported: Mode specific nodal displacement response missing in OptiStruct, not supported in NASTRAN template DRESP2 referring DRESP2 supported in OptiStruct template but not in NASTRAN template (identical syntax) DVPREL for PBUSH missing in NASTRAN template SET definition for DRESP entries missing 19/22

19 Definition of a complex optimization process Applicable for a wide range of structures Enabled by HyperWorks using tcl programming But there is room for improvements: More consequent support of NASTRAN keywords DRESP2 (just update the template) DVPREL (add further property types: PBUSH, ) SET (add further types: DRESP, ) 20/22

20 Q & A Thank you! Contact TECOSIM Technische Simulation GmbH Dr.-Ing. Martin Müller-Bechtel Tech. Manager Virtual Benchmarking Ferdinand-Stuttmann-Straße 15 D Rüsselsheim Phone +49 (0)6142 / Fax +49 (0)6142 / Mail m.muellerb@de.tecosim.com 22/22