THE FAULT DIAGNOSIS METHOD BASED ON JOINT APPROXIMATE DIAGONALIZATION OF EIGEN-MATRICES AND STACKED AUTO-ENCODER

Yuanfang Xin,∗ Yuanyuan Jiang,∗ and Yanbin Liu∗∗

Keywords

Joint approximate diagonalization of eigen-matrices (JADE), fault diagnosis, softmax, stacked auto encoder (SAE)

Abstract

To solve the problems of multiple soft fault modes of switching converter, such as cross or overlap and low diagnosis rate, a fault diagnosis method based on joint approximate diagonalization of eigen-matrices (JADE)– stacked auto encoder (SAE) is proposed. The main idea of fault diagnosis based on JADE–SAE proposed in this paper is as follows: firstly, the signal at the measurement point of the switch converter is compressed and preprocessed using com- pression sensing. In the compressed domain, the high-dimensional time-domain characteristic parameters of each measurement point’s signal are extracted and JADE dimension reduction fusion is carried out. These parameters are used in training the SAE network; thus, an improved network structure is established. Finally, the essen- tial characteristics of the circuit fault are mined, and the softmax classifier is used for fault diagnosis. The proposed method realizes efficient and fast identification of multiple fault modes of the switch converter.

Important Links:

Go Back