Application for fault diagnosis of loopers based on evolutionary KPCA-LSSVM

In this paper, an evolutionary hybrid approach is studied for fault diagnosis and it is applied to classify the loopers faults in hot rolling process. The algorithm called evolutionary KPCA-LSSVM is the combination of genetic algorithm (GA), kernel principal component analysis (KPCA) and Least Squares Support Vector Machine (LSSVM), which can obtain better fault recognition rate. Firstly, kernel function concept is introduced, and then GA is used to select the kernel parameter in order to improve the performances of nonlinear feature extraction and fault classification of KPCA-LSSVM method. Secondly, KPCA is used to extract the nonlinear principal features of loopers by adopting the optimal kernel trick to map nonlinearly the data into a feature space and employing the PCA procedure. Thirdly, the nonlinear principal features of loopers are taken as input into a LSSVM to classify the faults of loopers in hot rolling process. The results of contrastive experiments show that the evolutionary KPCA-LSSVM using GA to optimize the kernel parameters can extract fault features associated with the loopers effectively, reduce the computational cost and enhance fault classification properties.