Fault diagnosis based on deep learning

As representation scheme can severely limit the window by which the system observes its world, deep learning for fault diagnosis is put forward in this paper. It is a real time online scheme that can enhance the accuracy of detection, classification and prediction, and efficient for incipient faults that cannot be detected by traditional statistic technology. A stacked sparse auto encoder is used to learn the deep architectures of fault data to minimize the loss of information. Experiment results show that the proposed method not only improves the divisibility between faults and normal process, but also exhibits a better performance on the accuracy of fault classification for the chemical benchmark, Tennessee Eastman Process (TEP) data.