A Fault Diagnosis Method of Rolling Bearing Based on Improved Recurrence Plot and Convolutional Neural Network

The recurrence plot (RP) method has been introduced into bearing fault diagnosis due to its capability of effectively analyzing nonlinear and nonstationary waveform signals in dynamic systems. However, the interference of noise increases the difficulty of RP-based fault diagnosis. To solve this problem, this article proposed a novel antinoise bearing fault diagnosis method based on improved RP and a convolutional neural network (CNN). First, different scales of approximation coefficients and detail coefficients were obtained and constructed for RP based on wavelet packet decomposition (WPD) on the vibrational signal. Meanwhile, redundant parts of each RP were removed according to its symmetry characteristics, and the remaining parts of these RPs were spliced into multiscale asymmetric RP (MARP) containing all coefficients. Then, a fault diagnosis model for rolling bearing was established with MARP as the input of the pretrained ResNet-34. Finally, the validity of the proposed fault diagnosis method was validated on the Paderborn bearing dataset. Experimental results showed that the proposed fault diagnosis method achieved an accuracy of 90% under Gaussian white noise with a signal-to-noise ratio (SNR) of above −6 dB.