Application of SVD and transfer learning strategy on motorfault diagnosis

A novel approach utilizing singular value decomposition (SVD) for feature extraction and transfer learning for classification is presented in this paper aiming to improve the motor fault diagnostic performance under various operating conditions. A discrimination index is designed to describe the difference between the vibration features which are extracted by the Hankel matrix in SVD. The main idea of transfer learning is to utilize selective auxiliary data to assist target data learning, where a weight adjustment is involved in the TrAdaBoost algorithm for enhanced diagnostic capability. In addition, negative transfer is avoided through the similarity judgment between the target and auxiliary data. Experimental study indicates that transfer learning improves the diagnosis accuracy in complex conditions as compared to the traditional machine learning strategy.