Robust predictive control of diesel airpath

In an automotive diesel airpath application, the controller is required to track reference values of intake manifold (boost) pressure and exhaust gas recirculation (EGR) rate. EGR rate is the ratio of the EGR outflow rate to the combined EGR and compressor outflow rates. The reference values for these output variables are determined by a high level controller based on the current engine operating condition characterised by an engine rotational speed and a fuelling rate. Perfect tracking of these references in a drive cycle ensures optimal driver demand responsiveness, fuel efficiency, and satisfaction of legislated emission limits. However, diesel engine airpath control is challenging owing to its nonlinear multivariable nature. Conventional approaches use look-up tables and proportional-integral-differential (PID) loops ignoring the cross-sensitivities that can affect the controller performance. Furthermore, these approaches do not guarantee satisfaction of operating and reliability constraints on manifold pressures and physical limitations of the actuators. On the other hand, the constraint handling ability of model predictive control (MPC) makes it an ideal choice of control architecture for constrained multi-input multi-output systems, such as diesel engines. However, efficient calibration of typical implementations of MPC is hindered by the high number of tuning parameters and their nonintuitive correlation with the output response. Additionally, modelling errors might lead to constraint violation and loss of recursive feasibility. This thesis proposes a robust switched linear time invariant (LTI) MPC architecture for diesel airpath that utilises multiple local controllers that require low calibration effort. In order to reduce the number of effective tuning parameters to aid rapid calibration, first, structural modifications are introduced to the traditional MPC formulation by constraining the outputs within exponentially decaying envelopes and an appropriate cost function parameterisation. Subsequently, a methodology based on sequential convex program (SCP) is proposed to estimate maximal disturbance set