Knock detection in spark ignited heavy duty engines: An application of machine learning techniques with various knock sensor locations

Abstract

Knock detection is critical to engine control as it prevents damage and ensures optimal performance. However, it still presents significant challenges, particularly in alternative combustion systems, due to its complex nature. This study compares two machine learning (ML) approaches (supervised and unsupervised) for detecting knock in a heavy-duty engine, using various knock-sensor setups. It focuses on using accelerometer data as the primary input, a widely used on-road sensor whose compatibility with ML approaches requires further exploration. The analysis includes One-Class Support Vector Machine (OC-SVM) and Convolutional Neural Networks (CNN). Both methods demonstrated their ability to detect engine knock effectively, achieving sensitivity levels over 80% compared to MAPO (maximum amplitude pressure oscillation) index. The sensor placement revealed different effects over the results according to the method used, whereas the number of sensors showed a minor influence on the outcomes. The advantages and disadvantages of each method are discussed.