Design Of Fault Detection And Diagnosis In Batch Process Based On Fisher Discriminant Analysis

With the development of modern industries in extension and centralization,industrial processes have become more and more complex and the number of process variables is also increased.The safety of industrial processes needs higher requirements.So the processing monitoring has become one of the most hot research areas in process control.This paper aims to study the method of fault detection and diagnosis in batch process.Multivariate statistical analysis techniques do not rely on the math model of system,and they show particular advantages in dealing with the high-dimensional and coupling data,so they have become a important way to solve the problem of fault detection and diagnosis.Therefore,the Fisher discriminant analysis method is used to realize the fault detection and diagnosis of batch process.In the thesis,some new methods of fault diagnosis of batch processes are proposed based on Fisher discriminant analysis(FDA).The main contents are as follows:(1)The multi-model FDA method is developed to diagnose fault.The unmeasured data in a batch must be estimated for MFDA in terms of on-line fault diagnosis,which improves the rate of misdiagnosis.The multi-model FDA method builds a model for each moment to solve the problem of MFDA.The method is applied to diagnose faults in penicillin fermentation process,the rate of misdiagnosis is lower than the MFDA method.(2)A recursive multi-model FDA method is developed to diagnose fault.If the failure data is inadequate when modeling with multi-model FDA method.The diagnostic performance will be reduced.The recursive multi-model FDA method depends on the advantage of recursion to make use of fault data and has no effect on the production process.Comparing the three methods in the process of penicillin fermentation,the results show that the method is superior to the multi-model FDA and MFDA in fault identification.(3)A window-based multi-FDA model method is proposed to diagnose fault.If some sample data are missing when modeling with multi-model FDA method.Model can not be established.The window-based multi-FDA model method relies on the windowed characteristic to focus on the existing sample data and missing sample data to avoid the problem that the model can not be established.A penicillin fermentation process is used to test the performance of fault diagnosis of the proposed method.The results show that the method is effective.Through above studies,a recursive multi-model FDA method and window-based multi-FDA model method are better than the multi-model FDA in fault detection and diagnosis of batch process.Correct diagnosis rate is gradually increased,so above method can be applied to the practical industry.