Research And Application Of Fault Detection Method Based On Data

With large-scale and complicate for the industrial process, it is necessary to monitor the changes of process and the fault information. Multivariate Statistical Process Control (MSPC) methods have been widely applied to solve the problem of fault monitoring. ICA has been developed in recent years. The advantages are that it doesn’t assume the process variables to meet the Gaussian distribution as Principal Component Analysis (PCA), and the higher-order statistics information can be used. KICA has developed based on ICA and it can solve the nonlinear problem of data. Then by constantly improved KICA, it can help industrial process detect faults. This dissertation develops the research based on the predecessor’s work. The main research contents are as follows:(1) As large-scale industrial processes get more and more complex, the number of variables gets more and more large and it has a strong correlation among variables, the result can possibly obtain many false alarms and missed alarms by simple Multivariate Statistical Process Control (MSPC) methods. To overcome the above problems, the Multiblock KICA (MBKICA) algorithm is proposed. Firstly, it can divide the variables into some blocks, deal with each block using MBKPCA and the whitening matrix can be got. By the whitening matrix, the whitening data can be obtained. Then whitening data in each block is detected using ICA method. The new method applies into the cold rolling continuous annealing process.The simulation result of the process monitoring shows this method can not only improve the effect of the fault detection, but also narrow the location of faults.(2) Multimode industrial process is that a production line have many working conditions and different working conditions is correspond to different types products. The traditional multivariate statistical analysis method is not considering to the correlation among many modes, so it may have wrong monitoring results. Here, a multimode KICA algorithm is proposed. The basic idea is that the similarity and dissimilarity of different modes are first analyzed. That is to analyze different types of correlations from the cross-mode viewpoint. The common part is the similar variable correlations over modes, and the specific part is the correlations which are not shared by all modes. Then it applies KICA to monitor faults. The new method applies into the cold rolling continuous annealing process. The simulation result of the process monitoring shows this method can not only judge which mode the test data is in, but also reduce the false-alarms and improve the effect of the fault detection. What’s more, the multimode KICA give a comprehensive mode analysis and understand of the industrial process.