Anomaly Vibration Classification for Condition Monitoring in Passenger Cars

Abstract: This research aims to investigate accurate classification of anomalous vibrations produced during vehicle operation. The main research questions are which data can be used as classification criteria, which models can better achieve classification effects, and how to address classification issues under different variables in a car. Existing literature primarily focuses on fault classification in rotating machinery and anomaly detection in large-scale railway vehicle suspension vibrations, with few studies examining and classifying suspension vibration data from commercial passenger cars.  In this study, we conducted experiments and discussions separately on two variables: vehicle speed and driving scenarios. The original data includes vibration data from the suspension of four vehicles, steering wheel rotation angle data, Lateral Acceleration (ALAT), Longitudinal Acceleration (ALGT), and real-time vehicle speed during driving. Relevant features were extracted from the original data in the time and frequency domains to serve as training data for the model. The thesis discusses the performance of supervised and unsupervised learning for such classification tasks. The research found that supervised learning significantly outperforms unsupervised learning, and that certain methods such as autocorrelation are sensitive to driving scenarios. This research provides reference opinions for the use of classification models in vehicle anomaly vibration detection and classification, and discovers valuable feature spaces and methods to improve classification results in conjunction with industrial data.