A pairwise graph regularized constraint based on deep belief network for fault diagnosis

Abstract

An enhanced intelligent fault diagnosis method is proposed based on pairwise graph regularized deep belief network (PG-DBN) model. In this novel framework, two different graph constraints are imposed on hidden layer of the Restricted Boltzmann Machine (RBM). The first graph constraint defines the representation of preserving the feature manifold structure in same class of the data and the second graph constraint defines the representation of the penalty of the feature manifold structure in different class of the data. The two graph constraints introduce feature manifold structure to RBM and make the extracted features contain more intrinsic information, which contributes to a better classification result. Meanwhile, the convergence and stability analysis of the proposed method is presented. Finally, the advantages of the proposed fault diagnosis model are evaluated by Tennessee Eastman process.