|Abstract:||Selective catalytic reduction or SCR is coming into worldwide use for diesel engine emissions reduction for on- and off-highway vehicles. These applications are characterized by broad operating range as well as rapid and unpredictable changes in operating conditions. Significant nonlinearity, input and output constraints, and stringent performance requirements have led to the proposal of many different advanced control strategies. Moreover, hardware and software proliferation is driven by changes in catalyst formulation and converter size with engine and application.
To address these challenges, this dissertation introduces the first application of model predictive control (MPC) to automotive SCR systems. The controller includes a physics-based, embedded model which is general enough to represent all known ammonia or urea based SCR catalyst formulations. The model is calibrated to flow reactor data, enabling a priori controller changes in response to changes in catalyst formulation or converter size. Computational efficiency is achieved through use of successive model simplification, analytical solutions, and a varying terminal cost function. The controller is augmented with a gradient-based parameter adaptation law to achieve robust performance. Novel features of the estimator include metrics for time scale separation and statistical uncertainty, and component-wise separation of error covariance to maintain robustness in the face of sensor noise and plant-model mismatch.
The end result is a generic, map-less SCR controller offering excellent performance and intuitive tuning. As such, it provides a vehicle for collaboration among catalyst developers, system integrators, and control engineers.