Robust Cluster-Based method for monitoring generalized linear profiles in phase I

Abstract

Profile monitoring is one of the new statistical quality control methods used to evaluate the functional relationship between the descriptive and response variables to measure the process quality. Most of the studies in this field concern processes whose response variables follow the normal distribution function, but in many industries and services, this assumption is not true. The presence of outliers in the historical data set could have a deleterious effect on phase I parameter estimation. Therefore, in this paper, we propose a robust cluster-based method for estimating the parameters of generalized linear profiles in phase I. In this method, the effect of data contamination on estimating the generalized linear model parameters is reduced and as a result, the performance of T^2 control charts is improved. The performance of this method has been evaluated for two specific modes of generalized linear profiles, including logistic and Poisson profiles, based on a step shift. The simulation results indicate the superiority of this cluster-based method in comparison to the non-clustering method and provide a more accurate estimation of the parameters.