Fault detection of automobile suspension system using decision tree algorithms: A machine learning approach

Abstract

The study aims to detect multiple faults that are exhibited by suspension system components during prolonged usage. Faults such as strut worn out, strut external damage, strut mount fault, lower arm ball joint fault, lower arm bush worn out and tie rod ball joint fault were considered in this study. A novel approach is proposed in the present study that involves vibration signals and machine learning techniques to identify various suspension system faults. Vibration signals were acquired for different fault conditions (as mentioned above) at three different load conditions by a specially fabricated experimental setup. Statistical features were extracted from the acquired vibration signals from which the most significant features were selected using J48 decision tree algorithm. The selected features were provided as input to the tree-based family of algorithms to determine the best in class classification algorithm for suspension fault diagnosis. The results obtained enumerate that the random forest classifier produces the best classification accuracy for all the load conditions (no load, half load, and full load) with values of 95.88%, 94.88%, and 92.01%, respectively. Finally, the performance of the proposed classification model is compared with other state-of-the-art machine learning classifiers.