A Diagnostic Procedure For High-dimensional Data Streams Based Missed Discovery Excessive Probability Control

Data collection technology has been improved significantly in recent years with the innovation of science and the development of technology.Instead of just monitoring a single data,a real-time monitoring of high-dimensional data stream(HDS)is required in medical,mechanical,industrial production and other fields.This makes statistical process control(SPC)face new challenges.Once abnormal production process occurred,fault diagnosis in SPC can identify the cause of offset after the system sent OC alarm signal.Therefore the fault diagnosis of high-dimensional data stream has become a research hotspot in recent years.In this study,multiple testing method is applied to fault diagnosis in SPC field,and an HDS fault diagnosis program based on missed discovery excessive probability(MDX)is proposed.Firstly,the high-dimensional data streams in the production process are described as a statistical model,and then the fault diagnosis problem is transformed into a multiple testing problem.Secondly,the solution of the multiple testing problem is found by controlling MDX at a suitable level,which is used as the decision rule of the fault diagnosis problem,and its feasibility is proved by theoretical deduction.In present study,Oracle and Data-Driven fault diagnosis programs are also proposed based on whether the parameters and statistical information of the original data were known.At the same time,in order to prove that the proposed procedure in this paper has higher accuracy intuitively,it is compared with the missed discovery rate(MDR)-based HDS fault diagnosis program theoretically and numerically.The study shows it can reduce the probability of system error,identify OC data stream in production process to a greater extent,and improve the accuracy of system fault diagnosis.Finally,the procedure is applied to a group of semiconductor manufacturing production data,and the fault diagnosis of the process is completed.The elimination of process anomalies reflects the value of the program in practical application.