Bearing fault diagnosis using time segmented Fourier synchrosqueezed transform images and convolution neural network

In this paper, a time segmented Fourier synchro-squeezed transform-based convolution neural network is proposed for the bearing fault diagnosis. The proposed method acquired seven time-frequency transforms of vibration signals from the indigenously developed experimental setup and Case Reserve Western University bearing vibration dataset. The obtained time-frequency transforms are constant-Q transform, continuous wavelet transforms, Fourier synchro-squeezed transform, fast kurtogram, instantaneous frequency transform, Wigner-Ville transform, short-time Fourier transform. The images of these transforms are applied as input to CNN selected with optimal parameters to evaluate the fault classification ability. The proposed method achieved 100 % bearing fault classification accuracy on the developed experimental setup and CRWU bearing vibration dataset. The proposed method achieved average classification accuracy of 99.12 % for 5 dB signal to noise ratio and classification accuracy of 95.81 % for-5dB signal to noise ratio in a computational time of 45 s.