Designing a hierarchical neural network based on fuzzy clustering for fault diagnosis of the Tennessee-Eastman process

摘要

This paper proposes a hierarchical artificial neural network (HANN) for isolating the faults of the Tennessee-Eastman process (TEP). The TEP process is the simulation of a chemical plant created by the Eastman Chemical Company to provide a realistic industrial process for evaluating process control and monitoring methods The first step in designing the HANN is to divide the fault patterns space into a few sub-spaces through using fuzzy C-means clustering algorithm. For each sub-space of fault patterns a special neural network has been trained in order to diagnose the faults of that sub-space. A supervisor network has been developed to decide which one of the special neural networks should be triggered. In this regard, each neural network in the proposed HANN has been given a specific duty, so the proposed procedure can be called Duty-Oriented HANN (DOHANN). The neuromorphic structure of the networks is based on multilayer perceptron (MLP) networks. The simulation of Tennessee-Eastman (TE) process has been used to generate the required training and test data. The performance of the developed method has been evaluated and compared to that of a conventional single neural network (SNN) as well as the technique of dynamic principal component analysis (DPCA). The simulation results indicate that the DOHANN diagnoses the TEP faults considerably better than SNN and DPCA methods. Training of each MLP network for the DOHANN model has required less computer time in comparison to SNN model. This is because of structurally simpler MLPs used by the developed DOHANN method.