A Hierarchical Deep Neural Network for Fault Diagnosis on Tennessee-Eastman Process

Abstract

This paper proposes a hierarchical deep neural network (HDNN) for diagnosing the faults on the Tennessee-Eastman process (TEP). The TEP process is a benchmark simulation model for evaluating process control and monitoring method. A supervisory deep neural network is trained to categorize the whole faults into a few groups. For each group of faults, a special deep neural network which is trained for the particular group is triggered for further diagnosis. The training and test data is generated from the Tennessee Eastman process simulation. The performance of the proposed method is evaluated and compared to single neural network (SNN) and duty-oriented hierarchical artificial neural network (DOHANN) methods. The results of experiment demonstrate that our method outperforms the SNN and DOHANN methods.