Research On Multi-stage Online Fault Monitoring And Fault Diagnosis Method For Batch Process Based On KECA

Batch process is ubiquitous in modern industrial production which,because of its high added-value,low-volume,multi-species,serialization and other characteristics,is widely used in the field of bio-pharmaceuticals,food processing,manufacturing,fuel and spices and is closely related to the human.Therefore,efficient fault monitoring and fault diagnosis are essential in batch process to ensure production safety and improve product quality.Owing that batch process is generally featured with non-linear,non-Gaussian traits with high complexity,multi-stage and other characteristics,we propose a multi-stage batch process based on MKECA line fault monitoring and FWA-GRNN online fault diagnosis method based on multi-stage batch process.The main contents are as follows:(1)Multi-stage division based on KECA batch processAiming at the multi-stage characteristics of batch process,an adaptive phase partitioning method based on kernel entropy component analysis(KECA)clustering algorithm is proposed.The original normal batch 3D data set is expanded along the time slice to obtain the 2D data set,and then the KECA clustering algorithm is used to cluster the sampling points each time.This paper uses the criterion function value to determine the final cluster group number to realize adaptation phase division in the batch process.(2)Multi-stage online fault monitoring based on MKECA batch processAfter dividing the batch process into several stages,the MKECA fault monitoring model is built in each sub-stage to realize fault monitoring.In each sub-phase,the data processed by the KECA algorithm has good angular structure properties,rendering it able to construct a new angular structure similarity statistic and calculate the control limit of the new statistic by the kernel density estimation algorithm,thereby realizing the on-line fault monitoring of the batch process..(3)Multi-stage online fault diagnosis based on FWA-SVM batch processAfter the batch process segmentation and online fault monitoring,the fault diagnosis model of batch process in SVM parameters is proposed by using fireworks algorithm(FWA)to realize online fault diagnosis.Firstly,the FWA-SVM diagnostic model is constructed in each sub-phase.The original fault data is preprocessed and then applied into the respectivediagnostic models for training optimization to construct the optimal diagnosis model.Then the MKECA batch process online fault monitoring model is used to monitor the fault and output the fault data by pre-processing them and applying them into their respective sub-phase fault diagnosis models for online fault diagnosis.This method combines the advantages of both intelligent algorithm FWA and machine learning SVM to achieve efficient fault diagnosis in batch process.(4)Multi-stage online fault diagnosis based on FWA-GRNN batch processIn order to further the study on fault diagnosis method in batch process and achieve more efficient fault diagnosis,a multi-stage online fault diagnosis method based on FWA-GRNN batch process is proposed.Firstly,the FWA-GRNN diagnostic model is constructed in each sub-phase,and the new intelligent algorithm FWA is used to optimize the smoothing factor of the generalized regression neural network(GRNN)to construct an efficient fault diagnosis model.The method combines the advantages of the new intelligent algorithm FWA and the neural network GRNN,and can realize efficient fault diagnosis of batch processes.In order to verify the validity of the FWA-GRNN diagnostic model,the method is compared with the FWA-SVM method.The experimental results show that the neural network GRNN is better for batch process fault diagnosis,This verified the feasibility and effectiveness of the method.