ETH Zürich

University of Zürich

“Optimization and Applications”

 

Dr.-Ing. Thomas Vietor

Modelling and Stochastic Optimization in the Preprogram Phase

of the Development of Passenger Vehicles

 

Keywords

Vehicle Engineering, Vehicle Attributes, Robust Design, Stochastic Optimization

 

1. Introduction

The development of a passenger car is a multidisciplinary task. The vehicle has to fulfill demands out of different attributes like vehicle dynamics, driveability, acoustics, thermal and heat management, safety, durability, crash and economics. Very often demands out of these areas are conflicting. One main problem is the variability of mechanical quantities responsible for the performance and customer perception of a car. To overcome this, the extension of the conventional deterministic oriented development process to a process which includes stochastic quantities is necessary. In this paper the current deterministic approach is described briefly. It is not possible to perform the extension in a single step. In a first extended version of the process the variability of material parameters is included. In further steps the formulation and solution of stochastic optimization problems for sub-problems is necessary. Finally the complete approach should fully integrate the variability of stochastic quantities. This paper outlines the importance of the stochastic model used for the design variables and the stochastic parameters.

 

2. Development Process

Vehicle system concepts (e.g. body structure, front- and rear suspension, powertrain mounting systems, etc.), which are selected in an early program phase, have significant influence on the attribute performance of the vehicle. It is almost impossible to solve attribute concerns resulting from selection of poor concepts in a later program phase.

 

3. Introduction of Scattering Design Variables

In the development process a number of parameters and design variables are scattering. In conventional approaches they are assumed deterministic. With increasing importance of reliability the deterministic approach has to be extended and at least the most important parameters and variables have to be modelled stochastically. These most important quantities have to be identified with sensitivity methods. Only with the limitation to the most sensitive quantities the models can be calculated. The key factors for the introduction of scattering design variables are:

·             stochastic data like geometry, dimensions, thicknesses and statical and dynamical stiffnesses of different materials

·             different stochastic models like gaussian, weibull, lognormal distributions

·             objectives like sound pressure, vibrations, safety, ride and handling, costs, manufacturing, assembly, package

·             solution strategies like RSM, First Order Second Moment reliability methods (FOSM) and different methods of  stochastic optimization like mean value Taylor methods

·             use of Monte-Carlo methods for large scale models and structural analysis methods without semi-analytical gradient calculation

·             use of Genetic Algorithms for stochastic optimization of large scale models

 

4. Solution Strategies for Stochastic Variables

In the presentation different methods to handle industrial applications will be presented and characterized.

 

 

5. Augmented optimization procedure and sequential realization

An optimization loop has been augmented by an Advanced First Order Second Moment Method (AFOSM) - procedure for calculating the reliability indices. By integrating the procedure, one obtains two interlocked optimization loops. The same linkage would occur with a nonlinear structural analysis that is also performed iteratively. The outer loop comprises the quasi-stochastic optimization described above, while in the inner loop the reliability indices and the sensitivities are calculated.

 

Other authors are developing optimization procedures with the application of evolution methods. These methods are currently being tested at practical large scale models. The use of AFOSM and the application of evolution methods is compared for different criteria like convergence, number of structural analysis and costs.

 

Address: Thomas Vietor, Ford Werke GmbH, D-50725 Cologne, Germany.

This paper is describing the status of a project with substantial contribution of Dr.-Ing. Axel Hänschke and Dipl.-Ing. Jörgen Hilmann, Ford-Motor Co.