Unsupervised Deep Learning for Anomaly Detection in Automotive Trucks: A Survey

Abstract

As the collection of process data in intelligent vehicles progresses, using this data in data-driven prognosis models will become increasingly relevant. Anomaly detection in sensor data plays a critical role in ensuring vehicle safety, reliability, efficiency, and to automatically identifying abnormal behavior. The different operating points and design variants of the trucks make a manual analysis with statistical methods or expert knowledge impossible. Difficult is that, in most cases, there are no labels for the data, and primarily, only normal behavior data with sporadic error cases are available. Clustering, unsupervised, one-class classification, and anomaly detection approaches appear promising. This survey paper explores the application of unsupervised deep learning techniques in sensor data collected from trucks. We review and analyze various approaches, discuss their strengths and limitations, and identify open research challenges.