A Method of Fault Diagnosis Based on DE-DBN

Abstract

How to improve the accuracy of industrial process fault recognition and the efficiency of algorithm training has been the focus and hotspot in fault diagnosis field. In this paper, deep learning is introduced into this field, and a fault diagnosis method (DE-DBN) is proposed by combining DE algorithm and DBN. First of all, we have established a DBN model, which can extract the effective features from the massive fault data and realize the Tennessee-Eastman (TE) process fault diagnosis; Then a set of hyper-parameters of the DBN model are learned by the DE algorithm, which is used for hyper-parameter initialization of DBN; At last, during the adjustment of DBN network weights, the weights are updated by DE algorithm using random deviation perturbation, which makes the optimized DBN network get better fault diagnosis effect. After a lot of experiments in TE process and compared with other commonly used methods, the result shows that DE-DBN method can effectively diagnose and recognize multiple faults from the original signal, and have high accuracy and efficiency of fault diagnosis.