Study On Support Vrctor Machine-based Fault Diagnosis In Tennessee Eastman Process

With the development of computer technology and the application of automation technology in the process of industrial production,the complexity of technical processes are continuously growing. Once the abnormal situation isnot timely treated, not only performance degradation, economicloss will be caused but also serious catastrophes. Therefore, the fault diagnosis technologyis essential to ensure the safety and reliability of the production processes. As for the complex industrial process which is difficult to be modeled using the first principals, Fault diagnosis based on the data has become an important research field. Support vector machines(SVM) has a better generalization capability, absence of local minima,strong robustness, andsparseness of solution. Therefore, it has a unique superiority in fault diagnosis.In this thesis, we mainly study SVM based fault diagnosis schemes in Tennessee Eastman(TE) process. The major works of the thesis consist of thefollowing parts:1. A research on SVM based fault diagnosis scheme is carried out, then this fault diagnosis method is compared with the one using the Partial Least Squares(PLS) algorithm.2. Aiming at improving the SVM based TE process fault diagnosis scheme,three common optimization methods, i.e., Particle Swarm Optimization(PSO), Genetic algorithm(GA), Grid search(GS), are studied.Then they are put into the SVM basedfault diagnosis scheme, and how the diagnostic performance of the SVM classifier changes is examined.After that, the original data are normalized and dimension reduced using the principal component analysis(PCA) to study how the diagnostic performance changes.Finally, an improved SVM based fault diagnosis method of TE process is proposed. Compared with other SVM based fault diagnosis schemes, the proposed one has the best accuracy and costs the least time.