optimisation and robust design for car body development · optimisation and robust design for car...

Post on 26-Feb-2019

216 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

Optimisation and Robust Designfor Car Body Development

Dr. habil. Fabian DuddeckReader for Computational Mechanics

f.duddeck@qmul.ac.uk

Technical University of Munich

May-08 AUTOSIM - Duddeck 2

Objectives of AUTOSIM1

• Integration of simulation into the product development process (PDP)

• Characterization of materials

• Establishing confidence in the use of simulation

1 Taken from AUTOSIM network bulletin issue 3 (April 2008)

May-08 AUTOSIM - Duddeck 3

Objectives of AUTOSIM (cont.)

• To facilitate the use of advanced simulation technologies (FEM, CFD, etc).

• To improve the quality and robustness of modelling and simulation.

2. Quality Assurance methods for CAE / car body development

1. Optimisation and robust design for car body development

May-08 AUTOSIM - Duddeck 4

AUTOSIM - Main Topics

• Integration:Optimisation strongly depends on it.

• Materials:Optimal material selection

• Confidence:Robustness, Quality Assurance

May-08 AUTOSIM - Duddeck 5

Optimisation and Robust Design for Car Body Development

• Solid Mechanics (FEM):Evaluation of methods and algorithms for topology, shape and multi-disciplinary optimisation for crash and NVH;

• Fluid Mechanics (CFD):EU project FLOWHEAD, FP7, QMULFluid Optimisation Workflows for Highly Efficient Automotive Development Processes.

May-08 AUTOSIM - Duddeck 6

Optimisation for Solid MechanicsTypesShape

Particular geometry of the structure

Gauge

Fine adjustmentof the structure

• Optimal material selection• Optimal geometry• Robust results

Topology

Principal conceptsfor the structure

• Early and series development phases

• Optimisation under uncertainty / lack of knowledge

Source: Duddeck et al.: 3rd M.I.T. Conf. on ComputationalFluid and Solid Mechanics, Cambridge June 2005.

May-08 AUTOSIM - Duddeck 7

Gauge Optimisation

Intrusion

Limit = 100 %

Mass

Initial mass

Δm = -19.3 kg

Dynam ics

Limit = 100 %

91.1 %

Torsion I

Non-effective approach.

Mass

Initial mass

Δm = +9.1 kg

Iterative optimization required:

Frontal im pact

Source: Duddeck et al. (2005)

May-08 AUTOSIM - Duddeck 8

Multi-disciplinary Optimisation

Crash disciplines

NVH disciplines

Mass

Initial mass

Δm = -13.5 kg

Multi-disciplinary optimisation:Integrated optimisation approach

0

20

40

60

80

100

Degree of optimizedfunctionality

%

Fron

tLa

tera

lR

ear

Dan

ner R

ear

Dan

ner

Fron

t

Dyn

amic

sSt

atic

s

BMW (2001-2005): Evaluation study (4 test cases) for ~20 different algorithms

Source: Duddeck et al. (2005)

May-08 AUTOSIM - Duddeck 9

Multi-disciplinary OptimisationLateral impactIIHS, EuroNCAP,

Pole, etc.

MDO: 14 days300-500 different

designs

Rear impact(high speed)FMVSS 301

StaticsBending, torsion etc.

Frontal impact(high speed)

EuroNCAP, etc.

Frontal impact (low speed)Danner test

DynamicsBending, torsion etc.

Rear impact(low speed)Danner test

Source: Duddeck et al.

(2005)

May-08 AUTOSIM - Duddeck 10

Optimisation Algorithms

Non-linear optimisation

Search strategies Gradient-based strategies

Deterministicstrategies

Stochasticstrategies

Quasi Newtonstrategy

Gauss Newtonstrategy

Jacobmethod

Simplexmethod

EvolutionaryStrategies (ES)

GeneticAlgorithms (GA)

Simulated Annealing (SA)

BFGSmethod

Sequentialquadratic

programming

DampedGauss-Newton

Trust regionmethod

LevenbergMarquardt

May-08 AUTOSIM - Duddeck 11

Surrogate Models

Different disciplines: Crash, NVH etc.

Evolutionary Strategies Response Surface Methods

Advantages Disadvantages

Appropriate for all disciplines

Self learning strategy

High number of design variables

Global search possible

High numerical effort

Parameter of the algorithm are important

Slow convergence

Advantages DisadvantagesFaster if the response surface (RS) is “good”

Not appropriate for all disciplines(crash!)

Step-wise progression recommendable

Statistical values for quality of RS

High numerical effort for higher number of design variables

Different RSM available (which?)

May-08 AUTOSIM - Duddeck 12

Example for Surrogate Models

Ratio between via RSM predicted valuesand real computed values

Discipline PolynomialRSM Kriging

KernelSmoother

104 %

100 %

77 %

Frontal impact 87 % 115%

Lateral impact 97 %

103 %

101 %

Rear impact 98 %

Source: Duddeck et al. (2005)

May-08 AUTOSIM - Duddeck 13

Example for Surrogate ModelsFE-Modell

Successive polynomial RSM with ~250 samples

Source: Duddeck et al. (2005)

May-08 AUTOSIM - Duddeck 14

Intelligent Method Selection

Definition of the project

Generation of the initial set of variants (DoE)Computation and evaluation of the variants

Automaticdecision rule

SelectionRecombination

Mutation

Termination ?

Evolutionary Strategy

No

Choice of themeta model

Termination ?

Surrogate Models

No

Analysis of robustness

End of project

Statistical analysis of the resultsValidation of the approximation model

Stepwiseoptimization New generation

Fitness

Intelligent switchbetween strategiesdepending on the

discipline

May-08 AUTOSIM - Duddeck 15

Shape Optimisation

• Methods without re-meshing:e.g. morphing;Only for small geometrical changes;

• Methods with re-meshing:For larger geometrical changes;Still a challenge;

May-08 AUTOSIM - Duddeck 16

Parametric Model

Points and Lines

Cross Sections

Joints & Beams

Welds, Adhesives

FE Meshes

Beads, Stamps, Ribs

Freeform Surfaces

Loading, Etc.© SFE GmbH 2008

May-08 AUTOSIM - Duddeck 17

Parametric Model

Max 1

Min 1

Max 2

Min 2

© SFE GmbH 2008

May-08 AUTOSIM - Duddeck 18

Parametric Model

© SFE GmbH 2008

May-08 AUTOSIM - Duddeck 19

Shape Optimisation Example

K Volz, B Frodl, F Dirschmid, R Stryczek, H Zimmer (2007): Optimizing Topology and Shape for Crashworthiness in Vehicle Product Development. Int. Automotive Body Congress (IABC), 17 – 19, June 2007, Berlin.

May-08 AUTOSIM - Duddeck 20

Shape OptimisationDesign Variables

Volz et al. (2007).

May-08 AUTOSIM - Duddeck 21

Shape OptimisationInitial and Final Configurations

Initial guesses

Final design Volz et al. (2007).

May-08 AUTOSIM - Duddeck 22

1. Topology – Shape Optimisation

Initial model forshape optimisation

Topology optimisation of the frontend

Passenger cabin

Volz et al. (2007).

May-08 AUTOSIM - Duddeck 23

2. New Body Concepts

Referencestructure

New bodyconcept

Volz et al. (2007).

May-08 AUTOSIM - Duddeck 24

3. Shape Optimisation

Volz et al. (2007).

May-08 AUTOSIM - Duddeck 25

4. Validation

Volz et al. (2007).

May-08 AUTOSIM - Duddeck 26

5. Robustness

Which criteria is robust?Which criteria should be taken for surrogate models?

M Thiele, M Liebscher, W Graf: Fuzzy analysis as alternative to stochastic methods – a comparison by means of a crash analysis. 4th German LS-DYNA Forum 2005, Bamberg.

May-08 AUTOSIM - Duddeck 27

5. Robustness

Source: Renault, Europ. Autom. Safety 2004

Frontalimpact

Rearimpact

Lateralimpact

Loadpaths for

crash

May-08 AUTOSIM - Duddeck 28

5. Robustness

T Streilein, J Hillmann (2002): Stochastic Simulation and Optimisation focused on the VW Phaeton, VDI-Conf., Würzburg, Germany

Scatter Experimentn = 4

Intrusion 18.64 %

Scatter Simulationn = 72

Intrusion 11.41 %

May-08 AUTOSIM - Duddeck 29

CFD OptimisationFLOWHEAD (EU FP7)

Fluid Optimization Workflows for Highly Effective Automotive Development Processes

Collaborative ProjectCall: SST.2007-RTD-1 Competitive product development

Coordination: J. Müller / F. Duddeck, Queen Mary, Univ. of London

May-08 AUTOSIM - Duddeck 30

FLOWHEAD (EU FP7)Motivation and Approach

1. Derive methods for optimisation and robust design for fluid problems for the automotive industry;

2. Cover the full range of applications;3. Integrate the methods into workflows

adapted to PDP of the automotive industry.

The approach is mainly based on adjoint methods combined with topology, shape and parameter optimisation including robust design techniques.

May-08 AUTOSIM - Duddeck 31

FLOWHEAD (EU FP7)Strategic Objectives (cont.)

Improving automotive PDP in the following fields:1. Internal combustion: reduction of emissions / fuel consumption;

2. Aerodynamics: reduction of drag, fuel consumption, CO2-emissions;

3. Interior and exterior acoustics: reduction of noise, improving comfort;

4. Cabin comfort and safety: effectiveness of heating, cooling, defrosting, demisting, etc.

5. Thermal management: effectiveness of engine cooling

6. Exhaust systems: increasing catalytic conversion efficiency, reduction of emissions

May-08 AUTOSIM - Duddeck 32

FLOWHEAD (EU FP7)Partner

May-08 AUTOSIM - Duddeck 33

FLOWHEAD (EU FP7)Work Packages

WP 7(Queen Mary)

ManagementDisseminationExploitation

WP 1 (Renault) Integration into the PDP

WP 2 (FE-DESIGN) Optimisation Workflows

WP 3 (Queen Mary) Adjoint Methods

WP 4(TU Denmark) TopologyOptimisation

WP 6(TU Warsaw) Robust Design

WP 5(TU Munich) ShapeOptimisation

Start: Sep 2008

May-08 AUTOSIM - Duddeck 34

Quality Assurancefor CAE Car Body Development

• Current PhD project (BMW, TU Berlin, Queen Mary)

• Rising complexity due to international collaborations and outsourcing

• Derive a quality process for CAE (NVH, crash, etc.)

• Please contact me for more discussion

May-08 AUTOSIM - Duddeck 35

Future Challenges

1. Topology optimisation for crashworthiness;

2. Shape optimisation with large geometrical changes (re-meshing);

3. Automatic transition between shape and topology optimisation;

4. Combined material and shape optimisation including a stamp-crash coupling;

May-08 AUTOSIM - Duddeck 36

Future Challenges (cont.)

5. Optimisation for CFD;

6. Optimisation for fluid-structure interaction;

7. Simplified and surrogate models to reduce numerical effort (current activity: pedestrian protection);

8. Intelligent hybrid optimisation approaches;

9. True MDO algorithms; integration of costs and other disciplines

May-08 AUTOSIM - Duddeck 37

Future Challenges (cont.)

10. Combined optimisation and robust design;11. Quality Assurance methods for CAE;12. Database on variation in experiments;13. Integration into current development

processes;14. Higher confidence in simulation results.15. Knowledge on the limits of simulation.16. Design under Lack of Knowledge

May-08 AUTOSIM - Duddeck 38

Thank you.

Contact:Fabian DuddeckQueen Mary, Univ. of LondonSchool of Engineering and Materials ScienceMile End Road, London E1 4NS, UK f.duddeck@qmul.ac.ukPhone: +44-(0)20 7882 3749

top related