Robust och Multidisciplinär Optimering av Fordonsstrukturer

2009-00314

Fordons- och Trafiksäkerhet Resultatkonferens - 2014

Project Partners

Principal applicant: Volvo Car Corporation

Project partners: Combitech ABAltair EngineeringEnginSoft Nordic ABDynamore Nordic AB

Academic partner: Linköpings Tekniska Högskola

Overall Project Objective

Find suitable methods for implementing robust and multidisciplinary design optimization in automotive product development process

Sandeeep Shetty

Robust design optimization

Scope Develop efficient methodologies to perform multiobjective robust and reliability-

based design optimization of large-scale vehicle structures

Investigation of approximate modelling techniques to reduce the computational effort of the optimization process

Implementation of developed methodologies into the existing product development process

Sandeeep Shetty

Different approaches to evaluate robustness and to perform non-deterministic optimisation have been studied

An approach to perform multiobjective reliability-based optimization and robust design optimization is presented and verified using a vehicle side impact crashworthiness application

An efficient reliability-based optimization using a combined metamodel and FE-based strategy is proposed and illustrated using industrial examples

Comparison between FE-based and metamodel-based robustness analysis has been performed

An approach to handle the discrete responses using metamodels is also presented

PhD courses – 60hp

Overall accomplishments

Sandeeep Shetty

Define problem

• Inputs and outputs• Select Objectives• Uncertainties quantification

DOE strategy

• Design of experiments Verification

Robust design procedure

Optimisation strategy

Estimation of the mean and standard deviation

Meta model

‘

Design evaluation

Select a optimum design

Verification

Sandeeep Shetty

Article -1Robustness-analysis

Comparison between FE-based and metamodel-based robustness analysis

Validation of metamodels

New metamodelling approach to handle discrete responses is proposed

Conclusion Computational effort is

minimised significantly by using meta models

Meta-model approach had acceptable accuracy compared to FE-based approach.

Article -2Non-deterministic optimization

Comparative study of deterministic and non- deterministic optimization

An approach to perform optimization of large- scale vehicle structural application is presented

Conclusion Presented metamodel-based

approach was found to be suitable for large-scale deterministic optimization

Further improvement in the presented approach is required in the case of non-deterministic optimization

Article -3Efficient Reliability-based optimization approach

An efficient reliability-based optimization method is proposed and validated using industrial examples

Conclusion Proposed method has better

accuracy and the method is computationally efficient

Articles

Sandeeep Shetty

Documented ResultsLicentiate thesis

S.shetty: Optimization of Vehicle Structures under Uncertainties, Licentiate thesis, Linköping university, Thesis No. 1643

Journal Papers

S. Shetty and L. Nilsson: Multiobjective reliability-based and robust design optimisation for crashworthiness of a vehicle side impact, accepted for publication in the international journal of vehicle design. S. Shetty and L. Nilsson: Robustness study of a hat profile beam made of boron steel subjected to three point bending, Submitted for publication.

Conference Paper

S.shetty: Efficient reliability-based optimization using a combined metamodel and FE-based strategy. published in proceedings of 4th International Conference on engineering optimization (EngOpt2014) 

Sandeeep Shetty

Ann-Britt Ryberg

Multidisciplinary design optimization of automotive structures

Scope

Find an efficient MDO process for large-scale applications that takes the special characteristics of automotive structural applications into account considers aspects related to implementation within an organization and product development

process

Outcome

Description and demonstration of an MDO process that is simpler than multi-level methods fits existing organizations better than sequential response surface methods (SRSM) and direct

optimization often more computationally efficient than direct optimization, SRSM and multi-level methods

Ann-Britt Ryberg

Work performed

Literature survey• MDO methods• metamodel-based

optimization

Technical report

PhD courses• optimization courses• solid mechanics courses• etc

75.5 hp

Comparison ofMDO methods• single-level methods• multi-level methods

Conclusion:

A single-level method + metamodels is often the best choice

Article 1

MDO process• description• demonstration on a

simple example

Conclusion:

The process is efficient,flexible, and suitable forcommon automotivestructural MDO applications.

The process fits existingorganizations and productdevelopment processes.

etc.

Article 2

MDO studies• different software• different sizes• different methods

Experience

Licentiate thesis

Ann-Britt Ryberg

MDO process Application example

Decision

Initiation

Design of experiments

Variable screening

Metamodel creation

Setup

Verification

load case 1

Design of experiments

Variable screening

Metamodel creation

Setup

Verification

load case n

…

Optimization

Step 1Define problem (load cases, objectives, constraints, and design variables).

Step 2Find important design variables.

Step 3Define DOE, run simulations, and extract results.

Step 4Build, check, and compare metamodels.

Step 5Find optimum solutions.

Step 6Check results with detailed model.

Modal analysis

freq_m1_modalfreq_m2_modal

Front impactv

tx05_mid_frontintr_mid_front

Roof crush d

forc_3_roofforc_max_roof

Side impact

v

intr_upper_sideintr_lower_side

SetupMinimize mass without degrading the disciplinary performances.

Screening25 15, 7, 11, 12 variables

DOEAcceptable accuracy 90, 42, 55, 48 simulations

MetamodelsRBF neural networks +Feedforward neural networks

OptimizationAdaptive simulated annealing

VerificationRBFNN: 8% mass red. (1 constr. viol.)FFNN: 12% mass red.

Ann-Britt Ryberg

Publications

Licentiate thesis LIU-TEK-LIC-2013:1Metamodel-based design optimization – A multidisciplinary approach for automotive structuresby A-B Ryberghttp://liu.diva-portal.org/smash/record.jsf;jsessionid=d0d8422fc5bf97e6f729a89c0b32?searchId=1&pid=diva2:601789

Technical report LIU-IEI-R-12/003Metamodel-based multi-disciplinary design optimization for automotive applicationsby A-B Ryberg, R D Bäckryd, L Nilssonhttp://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-84701

Article 1Multidisciplinary design optimization methods for automotive structuresby R D Bäckryd, A-B Ryberg, L NilssonSubmitted

Article 2A metamodel-based multi-disciplinary design optimization process for automotive structuresby A-B Ryberg, R D Bäckryd, L NilssonUnder revision

Phase II accepted and started Project number: 2014-01340Aim:• Take researcher from licentiate to PhD.• Continue development of models for industrial problems.• Industrial implementation of the result from earlier project.• Couple the two areas in a combined study and paper..

Futured work