Fault Diagnosis of Electric Two-Wheeler Under Pragmatic Operating Conditions Using Wavelet Synchrosqueezing Transform and CNN

Abstract

Electric vehicles (EVs) have grown in universality in recent years due to their specific advantages such as zero emissions, low noise, and remarkable efficiency. Moreover, these vehicles, especially two-wheelers, becoming an integrated part of the present transport system, and their safe operation must be considered for a successful run under practical operating conditions. In the present work, a two-wheeler EV simulator is designed and developed to perform experimental simulations on different pragmatic operating conditions at the laboratory level. Moreover, a convolutional neural network (CNN) steered with wavelet synchrosqueezing transform (WSST)-based scalograms is proposed for fault identification in an electric two-wheeler. First, raw vibration signatures are acquired from the in-wheel brushless direct current (BLDC) hub motor and the mid-drive switched reluctance motor (SRM). Thereafter, these signatures are decomposed and transformed into time–frequency representation using WSST. Then, a CNN model is used for further identification and classification of the various bearing defects in EV motors. The experimental findings showed that the suggested method could successfully diagnose the various bearing faults under the mentioned working conditions. In both the in-wheel BLDC hub motor and the mid-drive SRM, the maximum achieved classification accuracies are 97.14% and 97.78%, respectively, in the case of the full-speed condition. It is evident from the obtained results that recommended approach can be used as an effective diagnostic tool for electric two-wheelers to prevent unforeseen vehicle motor breakdowns.