Low Complexity Model Predictive Control of a Diesel Engine Airpath.

Abstract

The diesel air path (DAP) system has been traditionally challenging to control due to its highly coupled nonlinear behavior and the need for constraints to be considered for driveability and emissions. An advanced control technology, model predictive control (MPC), has been viewed as a way to handle these challenges, however, current MPC strategies for the DAP are still limited due to the very limited computational resources in engine control units (ECU). A low complexity MPC controller for the DAP system is developed in this dissertation where, by "low complexity," it is meant that the MPC controller achieves tracking and constraint enforcement objectives and can be executed on a modern ECU within 200 microseconds, a computation budget set by Toyota Motor Corporation.

First, an explicit MPC design is developed for the DAP. Compared to previous explicit MPC examples for the DAP, a significant reduction in computational complexity is achieved. This complexity reduction is accomplished through, first, a novel strategy of intermittent constraint enforcement. Then, through a novel strategy of gain scheduling explicit MPC, the memory usage of the controller is further reduced and closed-loop tracking performance is improved. Finally, a robust version of the MPC design is developed which is able to enforce constraints in the presence of disturbances without a significant increase in computational complexity compared to non-robust MPC. The ability of the controller to track set-points and enforce constraints is demonstrated in both simulations and experiments. A number of theoretical results pertaining to the gain scheduling strategy is also developed.

Second, a nonlinear MPC (NMPC) strategy for the DAP is developed. Through various innovations, a NMPC controller for the DAP is constructed that is not necessarily any more computationally complex than linear explicit MPC and is characterized by a very streamlined process for implementation and calibration. A significant reduction in computational complexity is achieved through the novel combination of Kantorovich's method and constrained NMPC. Zero-offset steady state tracking is achieved through a novel NMPC problem formulation, rate-based NMPC. A comparison of various NMPC strategies and developments is presented illustrating how a low complexity NMPC strategy can be achieved.