Automobile clutch systems represent pivotal components within vehicles that facilitate smooth gear shifting, optimal engine operation, and efficient power transmission. Ensuring the integrity of clutch systems is paramount for maintaining vehicle performance and safety standards. Consequently, developing robust fault detection methodologies is imperative for promptly identifying potential issues. This study investigates the application of a CATboost classifier with feature fusion to analyse vibrational signals from clutch systems. Vibrational signals, collected via specialized sensors across various clutch conditions, serve as the primary dataset for fault detection analysis. A comprehensive feature fusion approach, combining statistical, histogram, and Autoregressive Moving Average (ARMA) features, aims to enhance fault detection accuracy. By integrating these feature sets, the study gains insights into clutch system behavior under varying operational circumstances. The classifier successfully identifies five distinct faults: worn release fingers, fractured pressure plates, deteriorated pressure plates, loss of friction material, and distorted tangential strips. Each fault presents unique challenges, emphasizing the significance of accurate detection mechanisms. Results underscore the remarkable performance of the CATboost classifier, achieving 100% accuracy when combining ARMA and statistical features. Impressive accuracy rates of 98.889% and 97.50% are observed with alternative feature combinations. Five other machine learning models (Decision Stump, Hoeffding Tree, REP Tree, SVM, and Random Forest) were also trained on the best feature combination set and compared to CATboost, illustrating its superiority. These findings substantiate the efficacy of feature fusion in augmenting fault detection capabilities within automobile clutch systems. The study's outcomes highlight the potential for improving vehicle maintenance practices, reducing downtime, and enhancing overall automotive safety through advanced fault detection techniques. Future research could explore real-time implementation of these methods in vehicle diagnostics systems.
