Anomaly Detection During Durability Testing of Rear-Wheel Steering Using LSTM Autoencoder

Rear-wheel steering (RWS) system, which enhances vehicle driving stability, has undergone durability tests and evaluation to improve steering performance until recently. However, RWS durability testing lasts for more than a month, requiring significant resources for continuous monitoring. As an initial step of Prognostics and Health Management research on RWS durability test, this study establishes a data acquisition system and proposes an anomaly detection method for analyzing the degradation process of the equipment. The data measurement system was used to collect five types of data, and anomaly detection on the RWS durability test data was performed using a long short-term memory autoencoder-based anomaly detection algorithm. The proposed method is trained exclusively on normal data and directly utilizes raw vibration data without any preprocessing. Additionally, by computing reconstruction errors for each segment, significant variations in reconstruction errors within specific segments can be observed. The proposed approach achieved an F1 score exceeding 99% and was applicable to vibration, vibration displacement, and acoustic data using the same model architecture. This study is expected to lay the foundation for automating the durability test monitoring process, fault diagnosis, and Remaining Useful Life prediction research.