Data analysis techniques for fault detection in hybrid/electric vehicles

Abstract

The modern automobiles of today are not merely mechanical systems, but are a complex product made up of several electronic, electrical and software systems. A typical vehicle consists of 70–90 ECUs, each designed and integrated to perform a specific function. Due to the complex nature of the interconnected ECUs and vast amount of data available in each of the ECUs, it is tedious to manually inspect the complete spectrum of vehicle data available in every test trips, and there is a high chance that many critical anomalies occurring for very short time spans(in milliseconds) are overlooked. The paper proposes modern machine learning based techniques to process the hybrid/electric vehicle data and detect anomalies in vehicle behavior faster and more effectively.