Research On Fault Diagnosis Method With Learning-Oriented Case-Based Reasoning For Tennessee Eastman Process

Chemical production process plays an important role in the national economy. In the production process, once a fault occurs but cannot be ruled out and eliminated in time, it will affect the production, resulting in unnecessary economic losses, and even threaten the safety of personnel and equipment, so fault diagnosis and real-time monitoring of chemical process has important practical significance. Case-based reasoning(CBR), as a new problem solving and machine learning method in the field of artificial intelligence, has been widely concerned in the field of industrial process fault diagnosis. However, when using the traditional CBR method for fault diagnosis, the similarity measurement method based on distance is often used in case retrieval stage. Because of this measurement method has two key scientific issues, namely how to allocate the weight of each attribute reasonably and how to avoid falling into the trap distance, has not been completely resolved. Therefore, on the basis of the traditional CBR fault diagnosis model, using learning pseudo-metric(LPM) instead of the distance metric, a new CBR fault diagnosis model which contains LPM retrieval is constructed in this paper, and Tennessee-Eastman(TE) chemical process was used to carry out experimental study. The main work is as follows:(1) CBR retrieval method based on distance metric is difficult to allocate the weight reasonably and easy to fall into distance trap. Aiming at these problems, a LPM retrieval method based on BP neural network is studied, which is used to replace the traditional distance measurement method. In this method, the learning pseudo metric is defined, and a pattern pool which is used to training the BP neural network is obtained by restructuring the source case. A new LPM case retrieval model which lays the foundation for the establishment of the improved CBR fault diagnosis model is gained when the network approaches LPM criterions.(2) In view of the fault diagnosis of TE chemical process, an improved CBR fault diagnosis model is established based on the LPM retrieval method. The model includes four parts: LPM case retrieval, case reuse, case modification and case storage. First,the fault diagnosis case base is constructed by using the TE simulation model. Then, the LPM model is established by training the BP neural network, after that the similarity between the target case and each source case is measured through the LPM model and similar cases which are similar to the target case can be obtained at the same time. Case revision method based on practice is used to confirm and adapt the diagnosis results afterwards. At last, the adapted case is stored. After all the operations, an incremental learning process is achieved.(3) An experimental platform of TE chemical process fault diagnosis based on C#, MATLAB and Server SQL is developed. The operation interface of fault diagnosis system is compiled by C#, and the historical data is stored in Server SQL. Fault diagnosis programs in the MATLAB are called by C#, which makes it convenient to observe the performance of different methods. The platform provides a verification tool for further study of the fault diagnosis method of TE process.