RNMF-SVDD Based Fault Detection And Diagnosis

With the constant enlargement of the scale of industrial process and complexity increasing, effective fault detection and diagnosis is the key to assure the safe production, improve product quality and economic benefits. On the other hand, with the rapid development of computer technology, a large amount of process data have been collected. How to transform these data into valuable information to improve the fault detection and diagnosis performance becomes a challenge issue.As a result of modern industrial process increasingly large-scale and complicated, for many complex processes, industrial process data often has the characteristics:(1) data has nonlinear;(2) data obey non-gauss distribution;(3) data has non-integrity characteristic. Based on the traditional multivariate statistical methods (PCA, PLS) did not respond to the above3characteristics.According to the process data of non-integrity, this paper proposed a robust non-negative matrix factorization for dimension reduction method. Non-negative matrix factorization in the realization of data compression and dimensionality reduction, and has positive characteristic and matrix sparsity etc., show a good data interpretation and the good ability of dell with incomplete data. This paper proposed the RNMF algorithm is a variant of NMF algorithm, mainly solves the problem of incomplete data.According to the process data of nonlinear and non-gauss distribution characteristics, this paper presents a fuzzy support vector data description method. As one class classifier, which is a variant of support vector machine, can establish the normal data hypersphere realize fault detection and the establishment of various fault hypersphere to realize fault diagnosis. The paper introduces the fuzzy concept, and has established a higher classification accuracy of fuzzy S VDD algorithm for fault detection and diagnosis.The article proposed a RNMF-SVDD based fault detection and fault diagnosis method, first by using RNMF method for dimensionality reduction, and then use SVDD to establish the normal data hypersphere realize fault detection and the establishment of various fault hypersphere to realize fault diagnosis. Based on the TEP (Tennessee Eastman process), the experimental simulation results show that the fault diagnosis method based on RNMF-SVDD with compare to PCA, NMF based on multivariate statistical method has higher accuracy and data integrity has good inhibition.