Research Of Nonlinear Processes Monitoring Method Based On Data

With the development of sensor technique,storage technique and the complexity of modern industrial process,the monitoring data of industrial process is becoming more and more abundant.Therefore,the process monitoring based on data has been received continuous attention and development.And multivariate statistical process monitoring(MSPM)techniques has been widely applied to process monitoring and received great success in the monitoring of industrial processes.Fault reconstruction based on PLS has been widely used to fault diagnosis.However,the normal information and fault information may be not effectively separated by the fault directions have been extracted as the normal data has not been considered in this method.Moreover,the triaditional MSPM is applied under the condition that the production process hode a single operation.However,the modes of production process are also exchanged as a result of the change of external environment,the demand of product diversity and difference of raw material composition.This thesis develops the research based on the predecessor's work.According to the fault diagnosis problems on nonlinear process and multimode process,the main research contents are listed as follows:(1)Firstly,kernel concurrent projection to latent structures(KCPLS)method is proposed which can provide a complete monitoring model.Then,KCPLS reconstruction method is proposed as a fault diagnosis method.On the basis of traditional reconstruction method,the relationship of fault data and normal data is analyzed deeply to extract the fault directions that cause the out-of-control statistic for each fault.Reconstructing the fault data using the reconstruction model of the same fault type.If the current fault model can remove the fault information from fault data accurately and eliminate the alarming of statistic,the fault isolation is therefore performed.The simulation results of fused magnesium furnace industrial process prove the feasibility of this method.Experiment results of penicillin fermentation process show the effectiveness of the proposed method.(2)Aiming at the problem of multimode process recognition and diagnosis problem in non-Gaussian process,this thesis proposed a between-mode relative analysis reconstruction algorithm based on kernel independent component analysis(KICA).KICA algorithm is used to extract the independent components,and then find the relationship between different modes.According to the relative changes which are obtained by the proposed algorithm,each mode is divided into three parts which contain the increased part,the decreased part and the unchanged part.Meanwhile,the three parts are divided into principal subspace and residual subspace as the data which has a large variance may cover up the useful information of the data which has a small variance.At last,establishing the monitoring modes of each principal subspace and residual subspace.Space division of different model has been completed based on the above introduction.And then,the fault amplitude and fault feature directions are extracted to reconstruct the fault data to gain the normal information from fault data and the fault diagnosis is completed.The performance of the proposed method is illustrated by Tennessee Eastman Process(TEP).Comparing to the traditional multimode method,the experimental results show the advantage of the proposed approach.