A Novel Intelligent Diagnosis Method for Bearing Based on Fused-feature Images

Abstract

Bearings are commonly used rotary components in many power transmission occasions. Reliability of bearing plays an important role in whole equipment operation. Automatic and reliable fault diagnosis of this rotary component can enhance its reliability and minimize the cost of equipment maintenance. Aiming at this common goal, a novel bearing fault diagnosis method is proposed, which is inspired by signal processing and image classification techniques. First of all, vibration signals in time domain are processed into frequency spectrum and squared envelope spectrum to refine more features. Subsequently, three kinds of features are adjusted to two-dimensional data which are fused and transformed into red-green-blue (RGB) color image form later. This action can utilize artificial techniques to fuse more information and enlarge differences among different types of faults. Finally, a deep convolutional neural network (CNN) method is adopted to extract features of the composite images and achieve fault diagnosis. Experimental results show that the proposed method can achieve a high fault diagnosis accuracy more than 99% for bearing.