Diagnosis Research Of SDG And SVM In TE Process Reactor

With the rapid development of scientific technology, such as control theory and pattern recognition, fault diagnosis technology is also pushed towards to development at the direction of the intelligecemany. Scholars have made a in-depth research and gained great achievements.At present, many problems exsit in intelligent fault diagnosis that the deficiency to obtain enough fault samples, whether samples obtained can correctly describe the spread characteristics of fault for object and redundancy variables can be excluded through the fault analysis, which influence fault classification accuracy and restrict the popularization and application of fault diagnosis technology. Support Vector Machine and Signed Directed Graph provide a valuable path to solve this problem for fault diagnosis.This paper mainly according to problems of support vector machine need to solve in fault diagnosis and analyse different Kernel function used in fault classification for TE (Tennessee-Eastman) process reactor systematicly and parameters optimization problems. SDG is one of qualitative analysis fault diagnosis method, SDG model can predict TE system reactor cause-and-effect relationship among different variables, the the failure propagation path and fault source orientation, meanwhile, compatible variables are got as support vector machine imput characteristic variables.Therefore, SDG and SVM are efficient to shorten fault judgement time for TE process reactor, and are of great theoretical significance and application value to the fault recognition.The main contents completed as follows:(1) Anylyse SDG method and establish centrifugal pump level system SDG model by using the deviation state of measured data, and concludes the fault propagation paths, get compatible variables and locate fault source.In this paper, feature variables for fault classification are obtained by establishing TE process reactor SDG model, get SVM, which greatly reduced redundant variables inconnected to reactor and reduce classifier number for SVM.(2) In order to make fault resolution higher, fault recognition system based on SDG and SVM is established in this paper, and do the experiments by SVM in the conditions of three kinds of different kernel functions for fault classification, and determined the kernel function relatively suitable for the TE process reactor; optimize SVM parameter combination (C, σ) by he grid search algorithm, three groups of higher fault resolution corresponding parameters combination are determined by experiment and compared with other methods for TE process reactor fault classification, this method has better effect in the faults diagnosis.Combine SDG and SVM to fault diagnosis for TE process reactor, SDG’s completeness and clear failure propagation path have a positive effect to get the key variables and improve SVM fault resolution.