The Research On Fault Diagnosis Method Of Chemical Process Based On Sparse Principal Component

The advancement of scientific and technological level has promoted the gradual trend of large-scale,intelligent and complicated modern chemical process.The data-driven related methods are used to monitor the occurrence of fault in real time,which has become an important means to ensure process safety and obtain benefits.The traditional data-driven methods not only have limitations in data preprocessing,but also have poor results in processing real-time data,discovering potential faults,and identifying the causes of faults.Therefore,fully extracting the data information of the real-time process,improving the detection accuracy of different faults and accurately identifying the specific causes of the faults have become important problems to be solved urgently.In view of the main problems faced by the actual chemical process,Firstly,the bat algorithm(BA)is used to optimize the variational mode decomposition(VMD)to obtain better mode number and bandwidth constraint parameters.Secondly,while the threshold and moving window are used to adaptively process real-time data,a weighting strategy is proposed to highlight the impact of minor faults.Finally,the Bayesian network(BN)of analytic hierarchy process approach with the entropy weight(AHP-EW)and sub-module fusion is used to diagnose the cause of the fault.The specific research work is described as follows:(1)Aiming at the problems of actual chemical process data such as noise,complex correlation and non-linearity,the VMD is effective for data preprocessing under the premise of verifying.The BA algorithm is used to optimize the pre-defined questions of the VMD to solve its mode number and bandwidth constraints.(2)Considering the limitations of the sparse principal component analysis(SPCA)method in processing real-time data,the moving window is used to achieve the replacement of new and old data while retaining its relevance.And the threshold method is adopted to select the model in the window,which is effective of adaptive processing real-time data.Aiming at the impact of potential minor faults,a weighting strategy is proposed to give different weights to highlight the existence of faults,thereby improving the accuracy of fault detection.(3)Aiming at the key problem of judging the specific cause of the fault,it is proposed to use the BN based on the AHP-EW and sub-module fusion to diagnose the fault cause.Firstly,the system modules are divided according to the process reaction mechanism,and the process variables in each sub-module are selected.Secondly,the AHP-EW is used to calculate the relative weight of each process variable in order of magnitude,the structure of the node sequence is more accurate given the BN.Then the results of the training of each sub-module are merged to obtain a comprehensive training structure,and the path of the fault is determined based on the reverse reasoning to determine the specific cause.Finally,the fault in the Tennessee Eastman(TE)process is performed case analysis to verify the applicability and effectiveness of the proposed method.