Fault diagnosis with feature representation based on stacked sparse auto encoder

Abstract

A deep learning method for fault diagnosis is proposed in this paper. The stacked sparse auto encoder(SSAE) model with the theory of deep learning extracts deep feature representation from original fault data. Compared with traditional methods, SSAE is more efficient because of its deep architecture. The feature representation is used by a softmax classifier for fault detection and classification. The proposed method is experimented on Tennessee Eastman Process(TEP), a chemical industrial process benchmark, to demonstrate its practicality and effectiveness.