Background: The clutch is an indispensable component within the automotive system, facilitating the transfer of engine power to essential drive components including, the wheels through the intricacies of gear shifts. When a clutch system falters, it disrupts the seamless gear transitions and power transmission, rendering the vehicle immobile. Efficient fault diagnosis is a critical endeavour, as it not only ensures the continued reliability of the automobile but also acts as a pre-emptive measure against unwanted breakdowns.

Methodology: This research paper introduces a pioneering approach to condition monitoring, rooted in the realm of vibration analysis, specifically tailored to detect faults within the clutch system. In this methodology, a suite of auto-regressive moving average (ARMA), histogram and statistical features are meticulously taken out of the signals of vibration acquired during operation. The discernment of the most pivotal features is employed the J48 decision tree method. Subsequently, a lazy-based classification is performed, examining three distinct load settings (10 kg, 5 kg and no load) and six varying clutch conditions.

Results: The results reveal test accuracies for individual features which are as follows: for statistical features, 88.33% (no load), 90.00% (5 kg load), and 90.83% (10 kg load); for histogram features, 91.67% (no load), 94.17% (5 kg load), and 83.33% (10 kg load); and for ARMA features, 99.17% (no load), 85.00% (5 kg load), and 96.67% (10 kg load). To further enhance classification accuracy, a feature fusion strategy is adopted, where two or more features are combined to assess their collective impact. The combinations of features examined include stat + hist, stat + ARMA, ARMA + hist and stat + hist + ARMA. Based on meticulous experimentation, it was observed that, for all the load conditions (no load, 5 kg and 10 kg) the combination of stat + hist + ARMA with Local KNN model achieved 100.00% classification accuracy.