Monitoring and fault diagnosis of batch processes using multi-model fisher discriminant analysis

Although multi-way Fisher discriminant analysis (MFDA) has been successfully applied in monitoring and fault diagnosis of batch processes, it has to estimate the future measured data of the batch processes during online monitoring and fault diagnosis. The estimated values not exactly follow the dynamic process behavior and they lead to false detection. In this paper, a novel method of monitoring and diagnosis for batch processes, named as multi-model FDA, is presented. The proposed method not only overcomes the need to estimating or filling up the future unmeasured data from the current time to the end of the batch, but also directly identifies the assignable causes of process abnormalities. Therefore, more accurate diagnostic decisions are made via multi-model FDA for online fault diagnosis. The method is proved to be effective by the application to monitoring and diagnosis of a multi-stage streptomycin fermentation process.