Non-linear identification, estimation and control of automotive powertrains
Increasingly tight emission regulations put a pressure on control engineers to come up with improved engine control systems. The task is difficult, as it is desired to minimize complexity, cost and maximize reliability and performance, all at the same time. Fortunately, modern control techniques offer assistance in achieving these goals. This motivation resulted in a range of topics developed in this thesis. A modelling, estimation and fault detection theory is presented. The estimation theory is often used for the system identification, but its main application is the model-based filtering, so important in real systems. The real systems are subject to failures. A theoretical development of the fault detection algorithm for non-linear systems is presented. The emphasis moves then to the control algorithms design. The non-linear algorithms based on the state-dependent model structure are introduced. An extension of the state-dependent Riccati equation method with a future trajectory prediction is developed. Also, the non-linear version of generalized predictive control algorithm is presented. Optimality of solutions is analyzed and corrections to algorithms are introduced to preserve the optimality. The theory needs practical verification. The identification of the spark ignition engine is presented next. A datadriven system identification method is developed. It provides an accurate model for control design purposes. The predictive control algorithm design is presented next. A simple air-fuel ratio control as well as a full multivariable control system design, with a torque as an output, is introduced. Improved tracking and tighter air-fuel ratio regulation is achieved. The control system efficiency may be impaired by the system noise and the model uncertainty. For that reason the model-based estimation techniques are very important. It is demonstrated that not only the noise immunity, but also robustness is significantly improved when Kalman filtering methods are employed. Last important topic of fault diagnosis is then presented. Faults must be detected, isolated and identified to enable successful control system re-configuration.