Acoustic signal based detection and localisation of faults in motorcycles

Vehicles produce dissimilar sound patterns under different working conditions. The study approaches detection and localisation of faults in motorcycles, by exploiting the variations in the spectral behaviour. Fault detection stage uses chaincode of the pseudospectrum of the sound signal. Fault localisation stage uses statistical features derived from the wavelet subbands. Dynamic time warping classifier is used for classification of samples into healthy and faulty in the first stage. In essence, the same classifier classifies the faulty samples into valve-setting, muffler leakage and timing chain faults in the second stage. Classification results are over 90% for both the stages. The proposed study finds applications in surveillance, fault diagnosis of vehicles, machinery, musical instruments etc.