Stochastic optimization methods applied to engine tuning.

Summary This thesis deals with stochastic optimization methods applied to engine tuning, so that the engine consumes as little fuel as possible and respects the pollution standards in force. This thesis proposes two modofocations of the existing methodology at Renault and a new approach that breaks with the process currently used. The first modification consists in reformulating the optimization problem using fuzzy logic models and the second one uses a new stochastic optimization algorithm "Multistosch", for which different convergence results are demonstrated. The new approach is based on a dynamic test planning tool (trajectories in a plane), using functional quantization and, in particular, the one under constraints. In this framework, a new form of constrained distortion is presented, which we will try to minimize by stochastic optimization algorithms.