Research On Process Monitoring Method Based On The IC-Subspace KICA And Kernal Marginal Fisher Analysis Of Prior Knowledge

The large complexity of modern industrial processes makes the information of monitoring data become more rich.The process control requirements are getting higher and higher.,So the scholars pay more attention for the fault monitoring of industrial process.The monitoring method based on data driven has been paid more and more attention in the field of fault monitor,and many scholars have made remarkable achievements in the study field.For the nonlinear,nongaussianity and multi-stage features of batch processes,this thesis presents a batch process monitoring method based on multi-directional kernel independent conponent analysis of independent conponent based on IC-subspace separation;Since there are only a small number of labeled samples and a large number of unlabeled samples in the industrial process.In order to use empirical knowledge,this thesis presents a fault detection method based on feature extraction of improved kernel Marginal Fisher analysis and local preserving of prior knowledge.And the method can improve the accuracy of classification.In this thesis,the training data is processed by batch expansion in order to solve the problem of three-dimensional data of batch process.In this thesis,we put forward a IC-subspace kernel independent component analysis fault diagnosis method in order to solve the invisible fault which involving a small number of variables.In this thesis,the independent meta-information is used as the basis for subspace partitioning.So every subspace is different from others.So that we can obtain to local information better.After that,a kernel independent conponent analysis monitoring model is established in each subspace.Multi-model joint monitoring on the same time.So the information that easy miss can be captured.Then we make the output together to monitoring.We deal with data from global to local and from local to grobal,so the local information of hide is expansed.This method can effectively improve the fault detection rate.As we know that we can obtain the labeled data difficultly and the unlabeled data easily in the industrial process.In order to obtain better detection results,in this thesis,a small amount of information is combined with a large amount of unlabeled information for feature extracted.In this paper,we use the local structure preservat and marginal analysis the based on Fisher analysis extracte feature information.And add kernal techniques to deal with non-linear data.This method makes good use of empirical knowledge.And preserves the spatial structure of the data in the feature extraction.Make a good monitoring on the local information.This method can make fault classification more obvious by cluster of the same classification and transpire of the different classification.The method can detect the fault information well by experiment.After comparison with other methods,we can see our method are better.