Data-driven Anomaly Detection of Engine Knock based on Automotive ECU

Abstract

In the automotive industry, the study of internal combustion engines (ICE) has massively been studied to identify the occurrence of some failures, such as engine knock [1], [2]. The occurrence of this phenomenon on the engine directly affects the engine maintenance cost and longer engine life. The use of machine learning for failure detection is highlighted [3]–[6]. An investigation was carried out by performing experiments with a Renault Sandero car, collecting some sets of variables for batch analysis. In this paper, we use artificial intelligence techniques with a data-driven approach, more specifically, machine learning, to detect the phenomenon of engine knock. The investigation was conducted with a feature extraction classifier, AutoEnconder Dense and Convolutional, SVM, and Isolated Forest. Finally, the best result achieved was 81% considering a feature extraction classifier on the collection of variables defined.