This paper presents the application of machine learning classification algorithms to identify and classify different heat release rate (HRR) shapes to control the combustion for an optimal multi-mode low-temperature combustion (LTC) engine operation. Low-temperature combustion engine produces low nitrogen oxides (NOx) and soot emissions and offers high thermal efficiency. But high in-cylinder pressure rise rates limit the operating range of the LTC engine. Therefore, it is imperative to control combustion in the LTC engine for safe operation. To this end, the HRR traces for over six hundred engine operating conditions are classified using supervised (i.e., Decision Tree, K-Nearest Neighbors (KNN), and Support Vector Machines (SVM)) and unsupervised (i.e., Kmeans clustering) machine learning approaches to segregate different combustion regimes based on HRR shape. Kmeans clustering was not successful in classifying the HRR shapes. Among different supervised machine learning techniques, SVM has proved to be the best method, having an overall classifier prediction accuracy of 92.4% for identifying the distinct shapes using normalized HRR data. In addition, three classifiers have been trained based on the combustion parameters and control inputs. These classifiers are then used as scheduling variables to develop predictive models. A model predictive control (MPC) framework is developed to control multi-mode LTC engine on cycle-to-cycle basis. The MPC framework achieved the simultaneous reference tracking of combustion phasing (CA50) and indicated mean effective pressure (IMEP) while constraining maximum pressure rise rate (MPRR) below 8 bar/CAD.
Read full abstract