Using Improved Self-organizing Map For Fault Diagnosis In Chemical Industry Process

Nowadays, fault diagnosis for complex chemical industry process has attracted more attention, in which the effective methods for visualization of fault diagnosis and monitoring are still challenging. The self-organizing map (SOM) has the ability of preserving the topological structure of the original data, so it is used to realize the visualization of fault diagnosis and monitoring. However, with the development of the chemical process, the connection of all components is much closer than before. The simple SOM algorithm cannot meet the requirements. Facing with this situation, two improved SOM methods are proposed for fault diagnosis and monitoring in this thesis, in order to improve the effectiveness of the visualization. The specific content is studied as follows:(1) Fault diagnosis based on self-organizing map integrated with correlative component analysis (CCA). The CCA considers the correlation between variables and fault categories. It can extract fault classification information as much as possible. Therefore, based on the identified correlative components, the map result of the SOM is greatly improved.(2) Fault diagnosis based on self-organizing map integrated with Fisher discriminant analysis (FDA). FDA is an optimal dimensionality reduction technique, which maximizes the separability of the classes. After the feature extraction by FDA, SOM can distinguish various types of fault states on the output map. Apart from this, a new fault identification method based on the correlation coefficient is used to provides a reference for fault sources.(3) A case study of the Tennessee Eastman (TE) process is employed to illustrate the fault diagnosis and monitoring performance of the proposed methods. The results show that both of the two methods are efficient to improve the effectiveness of visualization and capable for real-time monitoring in complex chemical process.