Non-linear model-based predictive control of a low-temperature gasoline combustion engine

  • Nicht-lineare Modellbasierte Prädiktive Regelung eines Niedertemperatur-Benzin-Verbrennungsmotors

Hoffmann, Kai; Abel, Dirk (Thesis advisor)

Aachen : VDI-Verl. (2010)
Dissertation / PhD Thesis

In: Fortschrittberichte VDI : Reihe 8, Mess-, Steuerungs- und Regelungstechnik 1172
Page(s)/Article-Nr.: X, 167 S. : Ill., graph. Darst.

Zugl.: Aachen, Techn. Hochsch., Diss., 2010


Topic of this thesis is the development of a non-linear MPC for the lowtemperature gasoline combustion (CAI) in a four-stroke single-cylinder engine. This process must be steered without the actuator of the spark plug. Moreover, smallest changes in the ambient conditions move the self-ignition towards disadvantageous timings and a rough combustion. For actuating the combustion in the single-cylinder demonstrator, the valve timings of the electro-mechanical valve train and the direct injection are available. Controlled variables are IMEP and CA50, while the maximum pressure rise dpmax shall be constrained. Alongside the V-model, different linear and non-linear controllers are developed. The non-linear modeling of the state space as neural network allows for its fast linearization and implementation in a non-linear observer structure as well as for the determination of a linear state space model. Linear and non-linear Kalman filters are discussed in combination with identified disturbance models, and the most suitable linear and non-linear structure is selected. Various methods for MPC are presented and combined with the established observers to (N)MPCs. Just like the observers, these are tuned automatically. Suitable structures are extended for the implementation of the constraint on dpmax. In an HIL test they are benchmarked under the assumption of heavy disturbances of the engine’s ambient conditions. The potential for robustness and calculation speed is evaluated and compared to a gain-scheduled PIDcontroller which unveils the necessity of non-linear MPC for the control problem. Recommendations for methods for the control of the non-linear CAI-combustion subject to constraints on the process are given. These methods are demonstrated exemplarily by controller application to the real engine. The simulation model is cross-validated using measurements of a controlled experiment. For comparison, the PID-controller is simulated in an HIL test under the same ambient conditions as measured in the experiments. Thereby, the developed procedure is justified which allows for the rapid control prototyping based on identification measurements.