A three-stage cross-domain intelligent fault diagnosis method for multiple new faults

Abstract Recently, transfer learning (TL) has been widely investigated to tackle the cross-domain fault diagnosis issue in machinery, and most research works follow the same assumption that the diagnosis domain shares the same fault categories. However, due to the randomness and complexity of mechanical faults, the new fault modes usually occur unexpectedly in the actual scenarios. The emergence of new faults also presents severe challenges to TL. In response to these challenges, a three-stage cross-domain intelligent fault diagnosis method is presented in this article. First, partial domain alignment is achieved based on an improved target weighted mechanism, and an outlier identifier is constructed to automatically separate the new fault classes. Then, an unsupervised learning model with silhouette coefficients is built to determine the number of new fault categories. Finally, the simulation signals are further adopted to distinguish the specific fault categories. Sufficient experiments on axial piston pump and public bearing datasets validate that the proposed method can predict a number of new fault categories and identify specific fault categories. The results indicate that the proposed method outperforms the other methods and has promising practical applications in fault diagnosis with multiple new faults.