Detecting known and unknown faults in automotive systems using ensemble-based anomaly detection

The massive growth of data produced in the automotive industry by acquiring data during production and test of vehicles requires effective and intelligent ways of analysing these recordings. In order to detect potential faults, data from the in-vehicle network interconnecting vehicle subsystems is recorded during road trials. The complexity and volume of this data keeps increasing since the degree of interconnection between the vehicle subsystems and the amount of data transmitted over the in-vehicle network is augmented with each functionality added to modern vehicles. In this paper, an anomaly detection approach is proposed that (a) is capable of detecting faults of known and previously unknown fault types, (b) functions as an out-of-the-box approach not requiring the setting of expert-parameters and (c) is robust against different driving scenarios and fault types. To achieve this, an ensemble classifier is used consisting of two-class and one-class classifiers. Without modelling effort and user parameterisation the approach reports anomalies in the multivariate time series which point the expert to potential faults. The approach is validated on recordings from road trials and it could be shown that the ensemble-anomaly detector is robust against different driving scenarios and fault types.