Research And Application Of Pattern Classification Method Based On KCVA In Industrial Process Fault Diagnosis

Production safety is one of the most important guarantees of "Industry 4.0".Process monitoring and fault diagnosis technology is an important tool for achieving safe production in industrial processes,and is one of the basic technologies for manufacturing transformation and intelligent manufacturing.With the development of large-scale production equipment and the intelligent development of instruments and meters,a large amount of process data generated in the production process can be saved,which creates good conditions for data-driven industrial process fault diagnosis.In this paper,the following research is conducted on fault diagnosis in complex industrial processes:(1)A new pattern classification method is proposed for industrial process fault diagnosis.The nonlinear signal processing tool Adaptive Rank-Order Morphological Flter(AROMF)is used to construct the pattern classifier.The method completes the classification of faults and provides a new solution for fault diagnosis in industrial processes.(2)In order to avoid the inaccuracy caused by pattern matching directly on the high-dimensional noisy raw data,the Kernel Canonical Variate Analysis(KCVA)algorithm is used to extract the nonlinear and dynamic characteristics of the process data to form AROMF template signal;then design the KCVA-AROMF pattern classification method combining KCVA and AROMF,based on the iterative error distance defined by the Euclidean Distance(ED)for fault pattern matching,and in the Tennessee Eastman Process(TEP)has been simulated and verified on the benchmark platform.Compared with the traditional multivariate statistical method,KCVA-AROMF reduces the fault error rate,especially for the difficult to diagnose faults,the fault error rate is more obvious.(3)For the AROMF pattern matching based on Euclidean distance,the two sequences are equal in length and the feature points are strictly corresponding to each other.A DTW distance based on Dynamic Time Warping(DTW)algorithm is proposed to improve the iterative error distance and form KCVA-AROMF-DTW pattern classification method,through simulation verification experiments on TEP benchmark platform,the results prove that compared with PCA-AROMF,CVA-AROMF algorithm,KCVA-AROMF-DTW improved fault diagnosis accuracy;compared to KCVA based on Euclidean distance-AROMF algorithm,KCVA-AROMF-DTW algorithm improves the sensitivity of single diagnosis.