Online Mean Shift Detection in Multivariate Quality Control using Boosted Decision Tree learning

Abstract

The rapid development of communication technologies and information and online computers and their usage in processes of the industrial production have facilitated simultaneous monitoring of multiple variables (characteristics) in a process. In this work, we applied boosted decision tree ( DT_boost) and Monte Carlo simulation to propose an efficient method for detecting in-control and out-of-control states in multivariate control processes.In this work, four classifiers (methods) - χ_¯X^2, χ_(X_new)^2, DT_(χ^2 ), T_c– are used for detecting the process control states. Then, with converting detection results these four classifiers, the boosted decision tree is made and provides the ultimate result as the in-control or the out-of-control states. To show how the proposed model works and the superiority of this method over χ_¯X^2, χ_(X_new)^2, DT_(χ^2 ), andT_cmethods, we run it on a standardized trivariate normal process. To compare and evaluate the performance of classifiers, we used ARL functions and the evaluation measures including Accuracy (ACC), Sensitivity (TPR), Specificity (SPC), and Precision (PPV).The findings not only showed the superiority of the proposed method over the tradition Chi-square but also confirmed former results on the efficiency of decision tree for rapid detecting of mean shifts in multivariate processes in which data are gathered automatically.