Condition Monitoring of an All-Terrain Vehicle Gear Train Assembly Using Deep Learning Algorithms with Vibration Signals

Abstract

Condition monitoring of gear train assembly has been carried out with vibration signals acquired from an all-terrain vehicle (ATV) gearbox. The location of the defect in the gear was identified based on finite element analysis results. The vibration signals were acquired using an accelerometer under good and simulated fault conditions of the gear. The raw vibration signatures acquired from all the possible conditions of the gear train assembly were processed using the descriptive statistics tool. A set of descriptive statistical features were extracted from the raw vibrational signals. This study used a deep learning algorithm based on the tree family, which includes the decision tree, random forest, and random tree algorithms, to classify gear train conditions. Among the tree family algorithms, the random forest algorithm produced maximum classification accuracy of 99%. The decision rules were used to design an online monitoring system to display the gear condition. This study will help to implement online gear health monitoring in ATVs, ensuring the safety of drivers.