With the development of industrial technology,industrial processes usually have multiple production stages.Therefore,the study of multi-stages fault detection methods is of great significance.Multi-stage complex working condition data has non-linear and dynamic characteristics.Aiming at the characteristics of the above data,a fault detection method based on the combination of time-space local nearest neighborhood standardization and principal component analysis(TSNS-PCA)is proposed.Firstly,the nearest neighbor set of the sample is searched at the time level,secondly,the nearest neighbor sample set of the time nearest neighbor of the sample is found at the space level,and then the T² and SPE control limits of the monitoring model are calculated on the standardized sample set by the principal component analysis method.Finally,the corresponding statistical value of the sample to be detected are calculated and compared with the control limits to achieve the purpose of fault detection.The TSNS method can transform the complex nonlinear,dynamic and multi-stage data to the origin of coordinates and make the new data approximately to a single normal distribution,while maintaining the degree of deviation between outliers and normal points.A non-linear dynamic numerical example and penicillin simulation process are uesd to prove that the effect of TSNS-PCA method is significantly better than that of PCA,LNS-PCA,DPCA,WkNN,TSNS-LOF and other methods in dynamic multi-stage process monitoring.

Fault Detection Strategy Based on the Feature Space Reconstruction of LPP [J]. Journal of Shenyang University of Chemical Technology, 2023, 37(5): 472-471.