Fault Diagnosis for Batch Processes Using Multi-model FDA with Moving Window

For batch processes, a complete batch trajectory can be obtained until the end of its operation. Besides, there are not enough fault data used to build models in historical dataset. Those make it very difficult to diagnose faults for batch processes based on Fisher discriminant analysis (FDA). In order to solve those problems a multi-model Fisher discriminant analysis with moving window is represented. Using moving window, the date used to build FDA model are made of the data of the current time and past time. Therefore, the proposed method not only sufficiently utilizes the finite fault data, but also overcomes the need of estimated or filled up the future unmeasured values in online fault diagnosis. An industrial typical streptomycin fermentation process is used to test the performance of fault diagnosis of the proposed method.