a comparison between one-sided and two-sided arnoldi based model reduction for fully coupled...

22
Vehicle Engineering Group A comparison between one-sided and two-sided Arnoldi based Model Order Reduction techniques [MORe] for fully coupled structural-acoustic analysis. Oral Presentation Session at the 153rd Meeting - ASA R Srinivasan Puri, Denise Morrey Oxford Brookes University, Advanced Vehicle Engineering Group, School of Technology, Wheatley Campus, Oxford OX33 1HX, United Kingdom. Jeffrey L. Cipolla Principal Development Engineer, ABAQUS Inc. 166, Valley Street, Providence, RI 02909-2499, U.S.A

Upload: reducedordermodel

Post on 20-Jan-2015

1.632 views

Category:

Automotive


0 download

DESCRIPTION

 

TRANSCRIPT

Page 1: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

A comparison between one-sided and two-sided Arnoldi based Model Order Reduction techniques [MORe] for fully

coupled structural-acoustic analysis. Oral Presentation Session at the 153rd Meeting - ASA

R Srinivasan Puri, Denise Morrey

Oxford Brookes University,

Advanced Vehicle Engineering Group,

School of Technology, Wheatley Campus,

Oxford OX33 1HX, United Kingdom.

Jeffrey L. Cipolla Principal Development Engineer,

ABAQUS Inc.

166, Valley Street, Providence, RI 02909-2499, U.S.A

Page 2: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Contents

1. Problem Description.

2. Idea of Model Order Reduction (MORe).

3. Model Order Reduction: Moment Matching.

4. Moment Matching: One Sided Arnoldi (OSA).

5. Moment Matching: Two Sided Arnoldi (TSA).

6. Model Order Reduction: Computational Aspects.

7. Numerical Test Case & Results: ABAQUS Benchmark Problem.

8. Summary.

Page 3: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Problem DescriptionCompute pressure level at drivers ear location (automobile or an aircraft interior) under structural or acoustic excitation.

Classical fully coupled FSI Formulation [Zienkiewicz & Newton 19691, Craggs,19712] :

Fluid Matrix

Structure Matrices

Coupling term

Displacements

• The direct formulation is the most accurate method when spatially variable, frequency-dependent trim material damping exists.

• Unsymmetric Mass, Stiffness Matrix increases computational expense.

• Modelling the final trim parts and joints leads to very high mesh density, and results in huge computational time.

Pressures

=

+

+

000

00 Fspu

KaKfsKs

pu

CaCs

pu

MaMfsMs

1 Zienkiewicz, O. C., and R. Newton, 'Coupled Vibrations of a Structure Submerged in a Compressible Fluid,' Proceedings of the International Symposium on Finite Element Techniques, Stuttgart, 1969.2 Craggs, A, 'The Transient Response of a coupled Plate-Acoustic System using Plate and Acoustic Finite Elements', Journal of Sound and Vibration, 15, 509—528, 1971

Page 4: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Idea of Model Order Reduction

=

+

00

02 Fspu

KaKfsKs

MaMfsMs

ω

Msa Ksa Fsa2ω− +

=

Mrsa Krsa Frsa2ω− +

=

{ } ε+==

Vzxpu

,

,

,

.

• Projection to lower dimensional subspace: States

Reduced States

=pu

Ly T)(ω

)()( ωω zLy Trr =

• Undamped:

Page 5: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Model Order Reduction: Moment Matching• How to pick projection matrix [V]?

- Moment Matching: Expand transfer function via Taylor series.

- Moment Matching: Match first moments for the transfer function of the coupled system.

∑ ∑∞

=

=

−− =−=0

2

0

211 )()1()(i

i

ii

isasa

isasa

Ti smsFKMKLsH

Su and Craig, 19915: choose projection matrix [V] to be the Krylov subspace to provide moment matching property.

- Explicit moment matching is unstable. Therefore, implicitly match moments via Arnoldi process.

➔ Modal Approaches: Uncoupled and Coupled (Morand and Ohayon 19973,

Ohayon 20044) projection formulation : Need to solve eigenvalue problem.

3 Morand, H. and Ohayon, R. 'Fluid Structure Interaction', ed. 1, John Wiley and Sons Ltd, 1995, ISBN-13: 978-0471944591.4 Ohayon. R. 'Reduced models for fluid–structure interaction problems', International Journal for Numerical Methods in Engineering, 60, 139--152, 20045 T J Su, R R Craig Jr 'Krylov model reduction algorithm for undamped structural dynamics systems' Journal of Guidance and Control Dynamics 14 1311-1, 1991

Page 6: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Moment Matching: Krylov via One Sided Arnoldi

Algorithm:1: Complete set-up for SISO / SICO Arnoldi Process (R.W.Freund , 20006)

6 Freund, R.W 'Krylov subspace methods for reduced order modeling in circuit simulation' Journal of Applied Mathematics 123 (1-2); 395-421, 2000.

Page 7: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Moment Matching: Krylov via Two Sided Arnoldi

Algorithm:2: Two-Sided Arnoldi Process (Grimme, E.J. 1997 7 and Salimbahrami, B. 2005 8)

7 Grimme E.J. 'Krylov Projection Methods for Model Reduction', PhD Thesis, Dept. of Electrical Engineering, University of Illinois at Urbana Champaign, 19978 Salimbahrami, B. ' Structure Preserving Order Reduction of Large Scale Second Order Models' PhD Thesis, Dept. of Electrical Engineering, Technische Universitaet Muenchen, 2005

➔ Accuracy of the one-sided projection can be improved by using appropriate two-sided techniques where twice the number of moments are matched.

➔ Output explicitly participates in the order reduction. Therefore SISO is restricted strictly to SISO.

Page 8: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

MORe: Modal Projection and Moment Matching

Table 1 – Comparison between coupled response prediction techniques

9 Everstine, G. C.. 'A symmetric potential formulation for fluid structure interaction' Journal of Sound and Vibration , 79, 157—160, 1981

Page 9: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Numerical Test Case: Benchmark Problem➢ABAQUS Benchmark Problem : 'Acid-Test' 10 :1.10.2 Analysis of a point-loaded, fluid-filled,spherical shell.

Model Description:•The model is a semicircular shell and fluid mesh of radius 2.286 m.

• A point load on the symmetry axis of magnitude 1.0 N is applied to the shell.

• The shells are 0.0254 m in thickness and have a Young's modulus of 206.8 GPa, a Poisson's ratio of 0.3, and a mass density of 7800.0 kg/m3.

•The acoustic fluid has a density of 1000 kg/m3 and a bulk modulus of 2.25 GPa. ● The response of the coupled system is calculated for frequencies ranging from 100 to 1000 Hz in 1 Hz increments.

➔ Obtaining accurate solutions in this case requires that the resonances and modes be modeled very accurately10.

Results for comparison:

Modal Expansion results from Stepanishen, P. and Cox 200010 : Compares results from Modal Expansion and ABAQUS Direct and modal projection solutions.

10 Stepanishen, P. and Cox, L. 'Structural-Acoustic Analysis of an Internally Fluid-Loaded Spherical Shell: Comparison of Analytical and Finite Element Modeling Results' NUWC Technical Memorandum, 2000 , Rhode Island: 00—118, USA

Page 10: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Numerical Test Case: Benchmark ProblemABAQUS Benchmark Problem : Simulated in ANSYS for initial comparison

Figure:4 : ANSYS Axisymmetric structural (left) and coupled (right) FE Mesh.

21907 Elements – Combination of ANSYS PLANE42 and ANSYS FLUID29 elements

Page 11: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Benchmark Test Case: Initial Comparison

Figure:5 : Driving point didplacement [Log]

Modal Expansion Results (Velocities) also Available Online: http://sufi.nchc.org.tw:2080/v6.5/books/bmk/default.htm

ABAQUS Benchmark Problem : ANSYS Comparison with closed form (Modal Expansion) Solution.

Page 12: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Benchmark Test Case: Order Reduction via Arnoldi

Figure:6 : ANSYS and Arnoldi predicted Driving point displacement

➔ Comparison between ANSYS and MORe via one-sided Arnoldi (OSA)

Page 13: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Benchmark Test Case: Order Reduction via Arnoldi➔ Predicted fluid pressure at the centre of the acoustic domain via OSA.

Figure:7 : ANSYS and Arnoldi predicted fluid pressure

Page 14: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Benchmark Test Case: Order Reduction via Arnoldi

Figure:8 : Local Error plot for fluid pressure: ANSYS and Arnoldi

➔ Local Error plot for fluid pressure at the centre of the acoustic domain via OSA approach.

Page 15: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Comparison between one-sided and two-sided ArnoldiABAQUS Benchmark Problem : Local Error plot for driving point displacement.

Figure:9: Local Error plot for driving point displacement - Comparison between one-sided and two-sided predicted results

Page 16: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Comparison between one-sided and two-sided ArnoldiABAQUS Benchmark Problem : True and Relative Error (Convergence) at start frequency

Figure:10 : Comparison between one-sided and two-sided convergence pattern.

Page 17: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Comparison between one-sided and two-sided ArnoldiABAQUS Benchmark Problem : True and Relative Error (Convergence) at end frequency

Figure:11 : Comparison between one-sided and two-sided convergence pattern for end frequency (1000Hz.).

Page 18: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Results: Computational Times: Benchmark Test Case

Table 2 – Computational Times; Benchmark test case

➔ Breakdown of computational steps for the Arnoldi based moment matching approach:

• Extract Matrices• Read Matrices and generate required (q) Arnoldi Vectors• Perform reduced harmonic simulation and convergence

Page 19: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Results: Computational Times: Benchmark Test Case

Table 3 – Split Computational Times – Benchmark test case

Page 20: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Results: Initial comparison with Modal approaches

Table 4 – Computational times for Modal projection techniques.

➔ Analysis carried out using ABAQUS V6.7. ➔ 20441 acoustic elements and 333 shell elements used for the Benchmark problem.➔ Frequency sweep: 100 to 1000 Hz.: 901 substeps.

Future Work:

➔ Accuracy comparison.➔ Current implementation for undamped/damped Arnoldi projection framework in Matlab/Mathematica.

11 ABAQUS V6.7 Theory Manual, ABAQUS Inc., USA

Page 21: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

Summary➔ The Arnoldi process generates orthonormal projection matrix [V] spanning the Krylov subspace to match the maximum number of low frequency moments of the coupled structural-acoustic system matrices – Resulting projected system is of much lower dimension.➔ Only matrix-vector dot product is required (+ 1 LU Decomposition).➔ Vectors are dependent on geometry (FE/FE Information), and can be efficiently incorporated into optimization or sensitivity analysis.➔ Complete output approximation is guaranteed for the one-sided Arnoldi process i.e. The Arnoldi generated matrix [V] can match both displacements on the structural domain, and sound pressure levels in the fluid domain.➔ Better approximation properties were found with the application of two-sided Arnoldi process. A MIMO version must be employed to match more than one specific output.➔ Extension to damped formulation can be made by the complex stiffness approach or by the explicit participation of [C]. Complex stiffness: Arnoldi; Future Work: Participation of [C]: First order transformation (or) Compute vectors using second order Arnoldi (SOAR) / two-sided SOAR .

Page 22: A Comparison between One-Sided and Two-Sided Arnoldi based Model Reduction for fully coupled structural-acoustic analysis

Vehicle Engineering Group

End.