Machine Learning-based Diesel Engine-Out NOx Reduction Using a plug-in PD-type Iterative Learning Control

Abstract

A plug-in Iterative Learning Controller (ILC) is proposed to reduce the engine-out Oxides of Nitrogen (NOx) emissions of a medium-duty diesel engine. A control-oriented model is developed to simulate the dynamic behavior of NOx, Carbon Monoxide (CO), and unburned hydrocarbon (UHC) emissions as well as engine power output given by the break mean effective pressure (BMEP). This control-oriented model consists of a support vector machine (SVM) that calculates the steady-state values of the emissions and BMEP as a function of the engine speed, the amount of injected fuel and the injection rail pressure. The SVM-based model was then augmented using experimental results from a fast response electrochemical NOx sensor to predict the transient behavior of the engine. Finally, a plug-in PD-type ILC that consists of a PID and an ILC controller is developed to reduce the amount of engine-out NOx while controlling the desired engine power, represented by BMEP, and monitoring the other emissions. The proposed controller provides a powerful tool for engine-out emissions trade-off in addition to controlling the desired engine output power.