Industrial Fault Classification Based On Multi-class SVM

This thesis studies the fault classification problems in complex industrial processes. Nowadays, industrial processes become large-scale, complicated and highly coupled. Any abnormal situations may be transmitted and amplified, that will lead to grave consequences, such as economic losses and casualties, to the whole industrial process. Therefore, fault classification problems in those processes are of great significance.So far, many data based fault detection and fault diagnosis methods have been well developed. Many multivariate statistical methods, including principal component analysis(PCA), independent component analysis(ICA) and Support vector machine(SVM) etc. have been proposed.In high-dimensional data classification, too many variables will lead to higher computation loads. Also, the noise in data lows the classification accuracy. Hence, data reduction also occupies an important position in the classification process. Nowadays, there have been many data dimensionality reduction methods, e.g., Principal Component Analysis, Kernel Principal Component Analysis(KPCA), Independent Component Analysis and Partial Least Squares(PLS) methods used in this thesis.In this thesis, Support Vector Machine and Principal Component Analysis based Support Vector Machine are used to classify faults. The classification accuracy is reduced due to Principal Component Analysis dimensionality reduction, and then Kernel Principal Component Analysis based Support Vector Machine is utilized to improve the classification accuracy. Kernel functions are applied, unknown parameters are introduced, and calculation process becomes more complicated during the Kernel Principal Component Analysis dimensionality reduction process. To avoid these problems, Independent Component Analysis based Support Vector Machine is used to classify faults. We found that Principal Component Analysis based Support Vector Machine and Kernel Principal Component Analysis based Support Vector Machine show poor classification performance on faults caused several disturbances simultaneously. After that, Partial Least Squares based Support Vector Machine are used to classify such faults. The traditional PLS category coding method cannot well reflect the correlation between categories. Therefore the classification coding method is improved, and the fault classification based on improved Partial Least Squares Support Vector Machine is proposed to gain a better classification result.