Bearing Fault Diagnosis Using Transfer Learning with ICCEMDAN

Abstract

Bearing is one of the fundamental components in the mechanical system and sometimes, it works for a long time under extreme conditions of high speed, which inevitably induces various faults. In recent years, many machine learning methods have been widely used in bearing fault diagnosis under varying working conditions. Currently, a major assumption in machine learning is that the training and test data must have the same distribution. However, it is very difficult to effectively obtain data that meets the conditions of independent and identical distribution of training data and test data, which may result in unsatisfactory fault diagnosis results. To solve the problem, a transfer learning fault diagnosis method with Improved Complete Ensemble EMD (ICEEMDAN) has been proposed. Firstly, the fault features in the original signal are extracted by ICEEMDAN, and a vibration signal can be decomposed into multiple IMF components. Next, the correlation value between the IMF component and the original signal is calculated to obtain the related IMF. At the end, the pretrained 1-D VGG16 model is trained by the target dataset, and the related parameters are fine-tuned. The iterated 1-D VGG16 model is applied for fault identification and diagnosis on the different distribution. Based on two online bearing datasets, transfer fault diagnosis comparison experiments under different working conditions have been carried out. The diagnosis accuracy of this method can reach up to 99%. Compared with 1-D VGG16 method and 1-D CNN method, the results show that the proposed method have better performance in fault diagnosis.