Multi-objective optimization through sequential validation with application to vehicle suspension design

The possibility of modeling complex phenomena without requiring full scale models for testing, has taken computer simulation to be a fundamental tool of modern engineering. Depending on the characteristics of the system to be modeled, different levels of detail can be identified, so it is important to assess the accuracy of computer simulations. Validation can be employed for assessing a numerical simulation through comparison with experimental data; taking into account errors and uncertainties and counting on a validation metric for quantitative assessment. A methodology has been developed previously in literature for addressing design optimization and validation. It consists in performing validation of designs as they are sequentially generated during a simulation based optimization process. Such methodology has been utilized with a single objective optimization problem. This thesis employs such methodology and a modified approach to it, with the scope of solving a multi-objective optimization problem which consists in finding the Pareto optimal set of two conflicting performances. A road vehicle suspension design problem has been considered. In addressing the multiple objective functions, the constraints method was used for scalarization. For validation, responses from the simulation model of the suspension are compared to the responses from a more accurate suspension model that has already been validated against experimental data. The confidence of the model is the validation metric; for its calculation, two statistical analysis techniques are used, which are probabilistic principal components analysis and interval based Bayesian hypothesis testing. The applied methodologies address also calibration, by maximizing the confidence at design points where the model is not valid.