Automotive is for me an almost inexhaustible source of complex problems to solve. Indeed, powertrain is a fast nonlinear complex system, multivariable, with saturated actuators. In addition, many important physical quantities are not measured and estimators/observers should be developed. Finally, the computation time is limited, and the control objectives (consumption, pollution, performance) are often competing, requiring a trade-off. This thesis summarizes my ten years of research activities on the control optimization of powertrain for automobile vehicles in three topics : Energy Management Strategies, Robust or Predictive control and observers. The common denominator of this work is to take into account nonlinearities and specificities of the process by seeking simplified knowledge models in order to implement controllers and observers in real time. The contributions include: a complexity reduction of optimal strategies by generic approaches with minimal optimality and the use of predictions ; a quasi-systematic implementation and design methodology of a robust or predictive control law from specifications for multivariable non-linear systems ; an estimation of unmeasured states by observers allowing high-level tuning between measurement confidence and model reliability ; experimental validation in the automotive domain of the proposed methods in order to minimize fuel consumption, pollutant and greenhouse gas emissions. My research project, which follows the logic undertaked so far combining automatic control and energetic, focuses on robust predictive control, on global multi-criteria and multi-variables strategies and on generic observers with applications in the automotive and energy fields.
