Mechanical noise identification using time-frequency representations

Three computation schemes of time-frequency representations (TFRs) have been developed and implemented to identify different components of mechanical noise originated from the transmission system of electrical vehicles. This study explores the close relationships between three TFRs, i.e. the spectrogram based on windowed Fourier transform (WFT), the Wigner-Ville distribution (WVD), and the smoothed WVD (SWVD). One main purpose is to pursue the efficiency of computing the SWVD of a dynamic signature. The revised scheme can tremendously reduce the computation time to a scale of around 1/90, compared with the original scheme. To assess the validation of these TFR schemes, firstly, four synthetic signals are designed and processed. Secondly, the developed TFRs are applied to distinguish different spectral components of transmission noise, and identify their sources. This study takes an electrical scooter with a continuous velocity transmission (CVT) system as a test bench. The CVT - belt noise, helical-gear whine noise, and fan noise can be clearly identified via the processing of the TFRs. These obtained conclusions can be used as references for machine element modification to improve annoying noise.