An Approach For Studying On Batch Process Fault Diagnosis Based On MFDA-PCA

Owing to the major shortcomings occure during the fault diagnosis on the batch processes that require pre-estimation all information arising after every current time by using traditional MPCA method, an approach for studying on batch process fault diagnosis based on MFDA-PCA has been proposed on the basis of PCA and FDA theories in this paper to cope with that problem, and then the method is applied to the on-line fault diagnosis for penicillin fermentation process. The whole process of fault diagnosis can be divied into two steps. First, the MFDA similarity analysis theory is adopted to build an optimal PCA model, and then the PCA model will be used to acheive the purpose of realtime monitoring for the whole online batch production process. Penicillin fermentation experiment this method verificates effective and rapid. The chief study of this thiese is showing as follows:1. On the basis of Fisher discriminant analysis (FDA) approach that used for pattern classification in its own characteristics of the maximum between-class scatter, an approach of similarity analysis based on MFDA method that is used between the process batch to batch has been presented in this paper, which can apply all the historical data for classification, and do similarity analysis between the tested batch data and the given batch data. Penicillin fermentation simulation results verify the correctness of the theory of similarity analysis, and lay the future theoretical foundation for on-line fault diagnosis process.2. Owing to the insufficient defect of online data that presents in the batch process fault diagnosis by using tradional MPCA method, an approach for fault diagnosis based on MFDA-PCA is proposed. This method first analyzes the online data by using MPCA similarity theory, and then uses optimal similarity batch to build a PCA model for real-time monitoring the new production batches, and completes diagnosis analysis for the appearead fault. Penicillin Simulation results show that the method is effective3. A set of batch process monitoring simulation system has been established based on an alcohol distillation theory and experimental conditions of a set of simulated production equipment. The monitoring system applys the configuration software to complete the entire on-line monitoring on production process and maintains the progress of production smooth. In the meanwhile, the system also provides data reporting capabilities to complete the function of query and save for the process variable data, especially has played a crucial role for the data-driven applications to those real engineering background.