Optimization tool dedicated to the validation process for the automotive industry

Abstract

Validation is a key process in the automotive sector. It ensures that the developed product fulfills the customer-oriented requirements at the system level. However, due to the increasing flexibility and complexity in the design of modern vehicles, this process has become more and more constrained and requires a tool dedicated to its automation and optimization.

This project intended to offer the basis of this optimization tool. In order to do so, it investigated two major parts of the tool: the interface dedicated to the validation and the optimization algorithm supporting the scheduling of the validation tests. This double study was the skeleton of this work.

First, a study of the validation process and project management theory was done in order to characterize the problem. It belongs to the discrete and non-linear category. Then, the potentially suitable optimization algorithms were reviewed and two of them were selected for the scheduling problem: the Genetic Algorithm (GA) and the Mixed Integer Distributed Ant Colony Optimization (MIDACO).

Mockups of the future interface were created, following the characterization of the validation process. Concerning the scheduling optimization, neither algorithm handled an increase of degree of freedom well, but MIDACO revealed to scale and to sample the solution space much better than GA did. Additionally, GA was intrinsically much slower and MIDACO provided fully reproducible results. The second algorithm was also more suitable for dynamic variations of the schedule.

In conclusion, the studied scheduling model can be used for further developments of a whole tool dedicated to the characterization, automation and optimization of the validation process. From the two studied algorithms, the MIDACO solver is the most suitable one and can be used for a more complex validation model, closer to the reality of operations or responding to the validation need of future complex developments.