Empirical study of features and classifiers for fault diagnosis in motorcycles based on acoustic signals

Motorcycles produce sound patterns with varying temporal and spectral properties under different working conditions. These sound signals carry important source of information which helps in automated diagnosis of faults. Fault diagnosis is a process of identifying the source of failure from a set of observed fault indications. This study gives an empirical analysis of features and techniques for fault diagnosis in motorcycles based on acoustic signals. The work proceeds in three stages: fault detection, faulty subsystem identification and fault localization. The time-domain, frequency-domain and wavelet-based features are considered for discussion. The features are tested with various classifiers at each stage of the experiment. Study reveals that the classification accuracy lies in the range of 70 to 100%. The proposed study helps in fault diagnosis of vehicles, machinery, tuning of musical instruments, and medical diagnosis.