Anomaly Detection in Electric Powertrain System Software Behaviour

Abstract

A software-in-loop (SIL) testing is a method of early testing of control software of a car in virtual environment. A system level testing is carried out on regular basis and it is important to see, if system is behaving as expected or unexpected. For unexpected behaviors, which test engineers not easily notice, modern techniques such as machine learning can give an advantage. This paper presents an application of machine learning algorithms that helps in identifying the abnormal patterns in time series data generated from electric powertrain system testing done in SIL environment for a Mercedes Benz Electric Car. Output of the SIL testing, results in time series data that is a collection of observations that are ordered chronologically and can be used to analyze trends, patterns, and changes over time. Anomaly detection in time series data is a process in machine learning that identifies data points, events, and observations that deviate from a dataset's normal behavior. By monitoring the expected and unexpected behavior of the electric powertrain system, anomaly detection can be a valuable tool for identifying potential issues. This study aims at coming up with an efficient process for anomaly detection in SIL. In order to get this process, various anomaly detection techniques are compared to detect a defined anomaly in time series data. Data pre-processing methods are also discussed before training the model. At the end, we conclude a best-fit method for identified anomaly. With finally identified method, a model was trained and used further in application. © 2023 IEEE.