Hierarchical Diagnosis Network Based on Easy Transfer Learning and Its Application in Bearing Fault Diagnosis

Abstract

Abstract Aiming at the problem of inconsistent distribution of rolling bearing vibration data under variable operating conditions, insufficient diagnostic data of the target bearing affects the accuracy of fault diagnosis, and the unknown severity of rolling bearing faults, a hierarchical diagnosis network based on easy transfer learning is presented in this paper and its application in the qualitative and quantitative diagnosis of rolling bearing faults. First, the wavelet transform is used to extract the fault features conducive to identifying the rolling bearing vibration data under various working conditions. Then, input the features extracted from the vibration signals of different fault types into the first layer easy transfer learning fault type recognizer to determine whether the target bearing is faulty and the fault type. After the fault type is determined, the features extracted from the vibration signals of the known fault types and different fault sizes are input into the second layer easy transfer learning fault size recognizer to determine the fault size of the rolling bearing. The proposed method is validated by the bearing data set of Case Western Reserve University and compared with other transfer learning methods that perform the same processing. The experimental results show the effectiveness and superiority of the method.