Integration of Pseudospectral Segments of Sound Signals for Fault Location in Motorcycles

Motorcycles generate dissimilar sound patterns under different working conditions. Generally, the experienced mechanics in authorized service stations take test rides to hear the sounds produced to diagnose the faults. The presented work attempts to locate the faults in motorcycles based on the sound. The features are extracted using pseudospectra of sound signals. Integration is performed for estimating the areas under different segments of the spectral curve. Manhattan distance is used for comparing the test feature vectors with reference feature vectors for classification. The results are over 85% for fault location experiment with combination of 4 faults and over 90% for combinations of 3 faults and 2 faults. The proposed work can be extended to consider some more faults of motorcycles. The proposed work finds applications in surveillance, fault diagnosis of vehicles, machinery, musical instruments and the like.