An acoustic approach for multiple fault diagnosis in motorcycles

Motorcycles produce sound signals with varying temporal and spectral properties under different working conditions. These sounds can be the source of information for automatic diagnosis of faults. Several models have been proposed for fault detection and diagnosis in motorcycles based on acoustic signals. However, these models are not capable of assessing multiple faults. This paper attempts to diagnose multiple faults in motorcycles using two-stage approach. During first stage the model identifies the vehicle is healthy or faulty. If vehicle is observed as faulty, the second stage identifies the major faults present. The distribution of energies, in the first five subbands of wavelet packet decomposition is used as features. A two-stage ANN classifier is deployed for recognition of faults in the fused fault signatures. The recognition accuracy is over 78% when trained with individual fault signatures and over 88% when trained with combined signatures.