Cross-Domain Fault Diagnosis of Bearings Using Simple Structure Model Combining with Signal Processing Method

Abstract

Bearings have essential roles in the operation and the safety of rotating systems. Artificial intelligence (AI) models have been developed to diagnose defects in various systems. They showed good performance for the trained system, and domain adaptation methods enhanced their performance for different operating conditions. However, those methods are not good enough when they are applied to the systems which have quite different characteristics from those of the trained system. In this paper, signal processing methods and the simple structure model are combined to diagnose different system from training system. The structural noise is removed, and one-dimensional convolution neural network (1D-CNN) combined with support vector machine (SVM) is used. The model is validated using published data (Case Western Reserve University datasets and Paderborn University datasets) and the results of domain adaptation method are also explained. The proposed method provides high accuracy of the cross-domain fault diagnosis even though only the normal data of target system are used in training classifier.