An Integrated Fault Identification Method Based on KPCA-PNN for Complex Industrial Process

When kernel principal component analysis(KPCA) method is applied to the on-line monitoring of complex industrial process,it is usually difficult to calculate kernel matrix K and to identify the primary fault sources.In order to overcome these bottleneck problems,an integrated fault identification method based on KPCA and probabilistic neural network (PNN) is developed.Firstly,a feature sample extraction method is adopted to pretreat the industrial data set.Secondly, Hotelling statistics T² and SPE of KPCA are adopted to detect system fault,and the gradient algorithm of kernel function is used to define two new statistics,namely C_T² and C_spe.Contribution degree of each variable to T² and SPE are calculated, and the fault features are extracted.Finally,PNN method is used to identify the primary fault sources from the features of correlative faults.The proposed method is applied to the simulation of Tennessee Eastman(TE) chemical processes,and the simulation results from several fault modes show that the proposed method can effectively identify various types of faults.