Research On Batch Process Fault Detection Based On Correlation Feature Analysis

As one of the main production methods in the development of modern manufacturing industry,batch process has the advantages of small production batch,variety,flexibility and so on.At present,it has played an important role in biological pharmaceutical,semiconductor,chemical energy and other manufacturing industries related to the national economy and people’s livelihoods.With the improvement of production automation,the production scale of batch process tends to be large and complicated.Therefore,the effective detection of the production process to ensure the safety of industrial production has become a hot issue of academic attention.A large amount of process operation data is transmitted and stored due to the rapid development of computer technology and sensor technology,which make data-driven methods use in batch process monitoring.However,batch process data are usually characterized by strong correlation,nonlinearity and uneven length.How to extract characteristic information reflecting production operation conditions from these process data is an urgent problem to be solved in batch process monitoring.To solve the problems of batch process,such as strong correlation,nonlinearity and uneven length between batches,based on the analysis of data association characteristics,this thesis proposes some corresponding batch process fault detection method.The main research contents are as follows:(1)A quality-related fault detection algorithm based on Neighbor Preservation Embedded Extreme Learning Machine(NPE-ELM)was proposed given the high complexity,strong correlation and nonlinear characteristics widely existing in batch processes.First,process variables were divided into quality-related subspaces and quality-unrelated subspaces by the Maximal Information Coefficient(MIC),and then batch process fault detection was carried out by NPE-ELM algorithm respectively.While reducing the dimension,the nonlinear and local neighbor structure of the data are kept and the accuracy of fault detection is improved.Finally,the effectiveness of the algorithm is verified by numerical examples and penicillin fermentation simulation experiments.(2)To solve the problems of weak interpretation ability of process variables to quality variables and nonlinearity in batch process fault detection,a method named Multi-way Orthogonal Signal Correction Enhanced Total Principle Component Regression(MOSC-ETPCR)is proposed for batch process quality-related fault detection.Firstly,3D data is processed into 2D data,the orthogonal signal correction algorithm is used to filter out the quality-independent information in the process variables,and the maximum information coefficient square matrix is introduced to extract nonlinear features and establish a regression model to ensure the maximum correlation between extracted features and quality variables.Secondly,the regression model is divided into quality-related subspace and quality-unrelated subspace,and the statistics and corresponding control limits are established respectively to detect quality-related faults.Finally,the effectiveness of the method is verified by a numerical simulation and penicillin fermentation process.(3)For the uneven length and key feature extraction of batch process,a fault detection and diagnosis algorithm for uneven batch process named Relaxed Greedy Time Warping Multi-way Orthogonal Enhanced Neighborhood Preserving Embedding(RGTW-MOENPE)is proposed.Firstly,in order to solve the problem of batch process unsynchronization,a relaxed greedy time warping(RGTW)method is used to synchronize the batch process data.Secondly,the enhanced objective function is constructed by obtaining the weight matrix.Then,the orthogonal basis function representing the local geometric structure is calculated,the key operation characteristics of batch process are extracted,and a statistical model is established for fault detection.After the fault is detected,the fault variables are diagnosed by the method of contribution graph.Finally,the effectiveness of the proposed algorithm is verified by penicillin fermentation process.