Research On Fault Monitoring Methods Of Single Batch Process Based On Phase Divide

Batch process play an important role in the modern process industry, social investment in this aspect is also increasing. Once an accident occurs, it not only can affect the reliable operation of the industry, but it even also affects the huge influence on our living environment. How to use multivariate statistical process monitoring method to carry on effective process monitoring of batch process, discover and eliminate the fault timely in the process, more and more getting the attention of the industry. This thesis analysis and comments the status of intermittent multivariate statistical monitoring methods systematically, combined the basic characteristics of the data with multi-period features of batch process, give priority to with single batch and conducts multi-period statistical modeling and the process monitoring research based on batch process.Firstly, the thesis introduced the intermittent process and its monitoring methods. On the basis of elaborating Principal Component Analysis (PCA), Dynamic Principal Component Analysis, Kernel Principal Component Analysis (KPCA) etc Multivariate statistical methods, analysis of the characteristics and the advantages of each method. Taking the dynamic and nonlinear characteristics of the data of single batch, emphatically analyzed the Dynamic Kernel Principal Component Analysis (DKPCA) which integrated from the above methods.Secondly, against that batch process tends to have periodicity, and the dynamic characteristics of data are different with different operating phase, the variables of the same operation phase are often highly nonlinear. For now mature modeling of multiple batches, some process data acquisition is difficult and the modeling data of multiple batches is large and complex etc problems, this thesis proposes two methods of batch process monitoring that are based on a single batch. First, in view of the batch process that the priori knowledge is rich, this paper proposed DKPCA process monitoring methods based on sliding window. In order to monitor the process accurately, conduct multi-phase divide modeling based on segmentation ideological. In the course of the method, on the basis of the single batch data, to set up the sliding window firstly, conduct DKPCA analysis for each window, and then compare the similarity of the load matrix between adjacent windows, the thesis puts that the similarity is greater than 0.85 classified as a same period, and achieved phase divide and modeling of batch process Secondly, for the batch process of no priori knowledge, this thesis proposed DKPCA process monitoring methods based GA optimization algorithm. In the course of the method, by optimizing the SPE to reach the minimum to determine the length of each period and KPCA kernel parameters, finally achieve the optimization DKPCA monitoring model of each period.At last, in the process of monitoring methods is proposed, penicillin fermentation process is used to verify both methods, the results show that the proposed methods are effective.