The Research On Dynamic Batch Process Monitoring Based On Improved Neighborhood Preserving Embedding Algorithm

Batch process,as one of the important production methods of modern manufacturing,is widely used in many industrial fields such as metallurgy,steel,pharmaceuticals,chemical industry and semiconductor manufacturing that are related to the national economy and people's livelihood,as well as emerging fields such as semiconductor manufacturing.However,once the production process fails,it will cause huge losses to the country and society,therefore,how to accurately and efficiently monitor the production status of the batch process and ensure the safe and reliable operation of the production process has be come the focus of attention.In recent years,with the development of computer technology and sensor technology,batch processes have accumulated a wealth of data reflecting process operation status and product quality in production,which have also prompted the rapid development of data-driven multivariate statistical monitoring methods.When the monitoring method based on multivariate statistics is applied to batch process monitoring,they often assume that the process is in an idealized state,that is,p rocess data obeys Gaussian distribution,process variables satisfy linear relationshi ps,single operating conditions and no outliers such as noise.However,the actual industrial process does not meet these assumptions,and because of the complex character istics of batch production in batch processes,how to establish a batch process monitoring model to achieve timely and accurate detection of batch process faults has become extremely challenging.Based on the neighborhood preserving embedding algorithm,this thesis proposes corresponding improved algorithms to achieve better monitoring effects due to the problem of poor process monitoring caused by outlier interference su ch as non-linearity,dynamics and noise in the intermittent process.The main research contents are as follows:(1)Aiming at the independent relationship and related relationship between batch process variables,a kernel dynamic latent variable-dynamic weighted support vector data description(KDLV-DWSVDD)is proposed based on variable blocks for batch process monitoring method.First of all,the introduction of mutual information can effectively evaluate the correlation and independent relationship between process variables,and divide the process variables into related variable sub-block and independent variable sub-block.Secondly,in the related variable sub-block,KDLV algorithm can effectively deal with the nonlinear effects of the process and extract dynamic information,in the independent variable sub-block,DWSVDD algorithm can effectively extract the dynamic information of independent variables,and at the same time,highlight the fault information by weighting.Finally,statist ical models are established in the sub-block of related variables and sub-block of independent variables for process monitoring.The simulation verification of penicillin fermentation process and semiconductor etching process prove the feasibility and effectiveness of the proposed algorithm.(2)Aiming at the dynamic relationship and static relationship between batch process variables,a dynamic-static joint indicator based on global slow feature analysis-global neighborhood preserving embedding(GSFA-GNPE)is proposed for process monitoring.Firstly,by evaluating the dynamic and statics of the pr ocess variables,the process is divided into two parts: dynamic variables and static variables.Secondly,in dynamic variables,GSFA can effectively extract the global and local dynamic information of the process.In static variables,GNPE can effectively extract the global and local static information of the process.Finally,a statistical model is established in two parts of dynamic variables and static variables to obtain statistics,and the statistics obtained from the two parts are combined by using Bayesian inference to obtain the joint indicators of the mixed model to realize process monitoring.Numerical examples and penicillin fermentation simulation process prove the feasibility and effectiveness of the proposed algorithm.(3)Aiming at the impact of information loss caused by three-dimension data expansion and noise and other outliers on process monitoring during batch process monitoring,a Markov chain neighborhood sparse preserving graph embedding based on tensor factorization is proposed for batch proc ess monitoring.Firstly,tensor factorization is used to directly process the three-dimension data in batch process,which can avoid the information loss.Secondly,when the nearest neighbor graph is constructed,the local geometric structure and sparse relationship of the process data need to consider,and at the same time,Markov chain theory is introduced,which can make the data after dimensionality reduction have a certain probability interpretation.Finally,a statistical model of the entire process i s built to monitor the process.Numerical examples and penicillin fermentation simulation process prove the feasibility and effectiveness of the proposed algorithm.