Process Monitoring And Fault Diagnosis Based On Kernel Entropy Component Analysis Methods

The automation control system of industrial process tends to be large-scaled and complicated constantly. Once the control system or equipment breaks down, it will be very difficult to indicate the fault with the single variable monitoring method. At this rate, it may lead to huge economic losses and even injury if the process ignores the failures. Therefore, besides the necessity to implement a good control strategy for industrial process, it is also important to study process monitoring and fault diagnosis methods.In the paper, the Tennessee-Eastman (TE) process is taken as the researching background and the process monitoring and fault diagnosis approach is studied based on data driven methods. It launches the research from the following aspects in the paper.(1) Based on the theory of Kernel Principal Component Analysis (KPCA), the theory of Kernel Entropy Component Analysis (KECA) and corresponding space transformation method are introduced. As a proof, it also discusses the differences between the KPCA and KECA through clustering tests.(2) Process monitoring strategy based on KECA and WT-KECA (wavelet de-noising method which is combined with KECA) are introduced. In the process of kernel mapping, Epanechnikov kernel function is used and the simulation test receives satisfactory results. That is to say, it can reduce the sensitivity of kernel parameter dramatically through introducing the kernel function. To some extent, the characteristic improves the application value of the KECA method.(3) A quality of the similarity of kernel entropy components space is defined and then a new fault diagnosis method based on the similarity is introduced in the paper. After comparing with corresponding experimental information, it proves that the method is efficient and reliable in the process of fault diagnosis.(4) A new process fault monitoring method based on TE subsystems is introduced. The method combines the knowledge of process with the corresponding strategy of system partition. The TE process is decomposed into independent subsystems in order to improve the sensitivity of the industrial process faults.The study result shows that it improves the fault detection rate with the KECA method and the variable contribution plots show the more outstanding fault location ability when compared with the KPCA method. The method of WT-KECA further improves the rate of fault-monitoring and the identification of fault variables. The fault diagnosis method based on the similarity of the kernel entropy components space fulfils some important fault identification tasks which are difficult to be completed by variable contribution plots. At last, the fault monitoring method based on the subsystems of TE process improves the fault detection ability further more. It also remains high rate of fault detection, especially from the aspect of applying to some hard-finding faults.