APPLICATION OF CONVOLUTIONAL NEURAL NETWORK AND CHAOS THEORY IN FAULT DIAGNOSIS OF ROLLING BEARINGS

Abstract

In order to solve the traditional methods in the process of judging the fault type of the bearing caused by signal strongly nonlinear misjudgment wrongly, based on the chaos theory, the phase space reconstruction method（PSR） is used to restore the system dynamics characteristics and the convolution neural network（CNN） is used to learn and extract the effective nonlinear information from the chaotic sequence, proposes the PSR-CNN intelligent fault diagnosis method, visualizes the attractor trajectory, and analyzes the nonlinear characteristics of each fault signal. Taking the experimental data of rolling bearings as the research object, the PSR-CNN method is used to analyze and diagnose early bearing faults. The results show that the attractor trajectory of the early weak fault signal is not representative of the fault due to noise interference; after learning by CNN and extracting effective nonlinear information, the attractor trajectory has significant chaotic characteristics and shows a fault separable state. The fault diagnosis method using PSR-CNN has higher accuracy and better generalization performance than the CNN diagnosis model based on the time domain and frequency domain, and has greater advantages in convergence speed and stability.

Keywords

• Convolutional neural network;Chaos;Bearing;Fault diagnosis