Fault Detection Of Batch Process Based On The Multistage MICA

With the progress of modern process industry, batch processes gradually plays an increasingly important role. Industrial production processes for safe operation and high standards of quality requirements have made the real-time online monitoring and fault monitoring research of batch process particularly important. This topic about characteristics of multistage and non-gaussian data distribution of batch processes, based on independent component analysis algorithm, researches a method of Fault detection of batch process based on the multistage MICA:(1) Adaptive phase partitioning algorithm of batch process operation phaseMore operation phases are the inherent characteristics of many batch processes. The standard FCM algorithm for phase partition of batch processes need to a given phase partition number beforehand, initialize clustering centers randomly, and is sensitive to noise and outliers. Hence, it puts forward the adaptive FCM algorithm with the application of clustering validity function to solve the above problems and achieve the adaptive partition of batch process operation phases.(2) Research on MICA algorithm based on particle swarm optimizationMulti-way independent component analysis has been long-term development in fault monitoring of batch process. Fast ICA algorithm as the common methods in ICA process monitoring model building research is vulnerable to the influence of the initial point when using, unable to converge to minimum point of gradient descent, and the independent principal component number is unknown beforehand. Therefore, the MICA based on particle swarm optimization algorithm is proposed in this paper.Independent component analysis is suitable for non-gaussian monitoring field, data characteristics lead to the difference in determination the confidence limit of monitoring statistics of PCA, PLS methods. This paper introduced support vector data description algorithm to determine the confidence limit of monitoring statistics, avoiding the "dimension disaster" problem caused by kernel density estimation.(3) Proposed the method of Fault detection of batch process based on the multistage MICAFor the actual industrial batch process, historical data often meets non-gaussian distribution and has obvious multi-stage features. So, this paper proposed the method of multistage MICA to implement effective monitoring of industrial batch process. Firstly, this method uses the adaptive FCM algorithm for phase division of batch processes. Then, the child MICA model optimized with PSO in each stage is established. Finally, use the improved statistics confidence limit for process monitoring. Industrial penicillin fermentation simulation experiment shows the proposed method make the process monitoring more timely and effective.(4) Field experiments of E. coli fermentation processIn e. coli biological fermentation process monitoring, the experimental results show that the multi-stage and non-gaussian data with the batch process, this method not only can reduce the false alarm at the normal operating process, and can reduce the omission of the failure process, to ensure that operators find fault in time.