Research And Realization Of Fault Diagnosis System In Process Industry Based On Improved Semi-supervised SVM Algorithm

The fault diagnosis of process industry has the vital significance to inhibit accidents and reduce economic loss, and the automation and intelligence of industrial process monitoring and fault diagnosis has become one of the hot research area in the industrial automatic control area. With the development of modern industrial detection methods, process data acquisition become more real-time and reliable. Because the fault diagnosis methods based on machine learning and artificial intelligence do not depend on accurate mathematical model, it cause a lot of attention by experts in the research area of process industrial fault diagnosis. At the same time, because of the complexity of the process industry, there are a lot of unlabeled state data. How to use these unlabeled data in industrial fault diagnosis causes more and more attention. The process industrial fault diagnosis method based on improved semi-supervised SVM algorithm in this paper is combined with machine learning, SVM algorithm in artificial intelligent area and semi-supervised algorithm for the application in the real-time fault diagnosis area.The principle of SVM was studied in this paper. The improved semi-supervised SVM algorithm proposed in this paper was combined with semi-supervised algorithm and SVM. The new proposed algorithm introduced unlabeled samples in the industrial process to the building process of SVM diagnosis model, proposed a new unlabeled sample selection system, solved the imbalance problem of diagnosis data, modified the SVM hyperplanes and improved the fault diagnosis rate.Considering the characteristic and needs of modern process industry, the proposed new fault diagnosis method-semi-supervised SVM fault diagnosis-was applied to the typical blast furnace smelting process and TE chemical process. Comparing with the other SVM fault diagnosis method through a lot of simulation experiment, the results proved the superiority of proposed new semi-supervised SVM diagnosis method in using unlabeled samples, improving the performance of classifier, and verifying the effectiveness and feasibility of the new method in process industrial fault diagnosis area.In the Windows XP environment, a blast furnace fault diagnosis system was developed using KINGVIEW 6.53, VB6.0 and SQL Server 2000. Considering the requirement of actual production, every function module of the system was designed particularly, and the blast furnace fault diagnosis system based on improved semi-supervised SVM algorithm was realized. The diagnosis system has rich function and strong practical applicability.