Kernel Fisher Envelop Surface Based Fault Diagnosis For Batch Processes

Due to less investment in equipment, short production cycle, the advantages of flexible operation, batch process is widely used in modern production. In order to meet the demand of the market and improve enterprise competitiveness, batch process increasingly tends to be the direction of efficiency, large and integration. At the same time, people also pay more attention to the safety and reliability of batch processes. Therefore, the study of fault detection and diagnosis for batch process is very important and meaningful. The complexity of the batch process increases the difficulty to establish mathematical model for diagnosis. Monitoring method based on data driven only depend on the collected data, and it is the research emphasis in this paper.Batch process inevitably exists the change of the initial conditions, external environment and the aging equipment can cause the change of the production cycle. However, in view of the batch process monitoring methods are generally requires equal production cycle. In addition, some methods in online monitoring need the complete production, but the fill and estimation the unknown values, this will lead to reduce diagnostic performance. For the above two problem, based on Fisher discriminant analysis (FDA) we proposed fault diagnosis based on kernel Fisher envelop surface for batch processes. This method takes advantage of the Fisher discriminant analysis to classify the category, and establish the envelope surface model for normal data and fault data respectively. Compared with multiway Fisher discriminant analysis, this method spread the data according to the direction of batch, can solve the problem of inconsistent production cycle, do not need the complete production and use the kernel function to deal with complicated nonlinear. By simulation comparison of the traditional kernel Fisher discriminant analysis and improve of the multiway Fisher discriminant analysis, it proved the effectiveness of the proposed method.In view of the Principal Component Analysis (PCA) in the modeling using only the data under normal working conditions, not considering the role of known fault information, and the FDA while using the fault class information, the fault diagnosis performance is better than PCA. According to the above problem, we proposed Fisher envelope analysis and PCA fusion fault diagnosis method for batch process, combining the effective fault detection ability of PCA and the outstanding fault diagnosis of FDA. At last, this method is verifies by penicillin fermentation simulation platform.