Anomaly Monitoring And Capability Analysis For Nonstationary Industrial Processes

Process anomaly monitoring and capability analysis are playing key roles to ensure production safety and product quality in intelligent manufacturing.In recent years,data-driven relevant methods have achieved great research results.However,most of the existing researches are for stationary process.With the manufacturing being more complex and delicate,the influence brought by nonstationary processes is becoming significant,that has been widely concerned by academia and industry.In nonstationary process,process variables are time-varying,which leads to the problem of complex feature extraction and asynchronous process;the corresponding quality characteristics may tend to be multivariate,non-Gaussian and nonlinear.These problems have brought a big challenge to process monitoring and capability analysis,remaining to be further researched.To solve the mentioned problems in fault detection,a modified DTW algorithm was proposed.Based on the neighbors measured by DTW distance and an open-ended projection strategy,the time axes were fixed for asynchronized data firstly.To detect time-scale faults and amplitude-scale faults,two monitoring indexes and the control limits of them were established then.The proposed method can effectively monitor nonstationary process.Through the cases of the Tennessee-Eastman(TE)process and a semiconductor etching process,the higher accuracy of the proposed method has been proved comparing to the existing methods.As for the multivariate capability analysis of controlled process,the quality data were augmented based on their distribution features through kernel principal component analysis(KPCA)and Yoe-Johnson transformation at first.Then,a visualization framework based on a modified Kiviat diagram was introduced,the confidence region of each quality grade was determined respectively to evaluate the products accurately.Finally,a new process capability index(PCI)was established to analyze the capability level of the process corresponding to multi-grade specification limits.The proposed multi-grade capability analysis method was tested through the products of a laser-semiconductor manufacturing,the results demonstrated the proposed method is more effective than the existing methods.