ROBUST CONTROL CHARTS BASED ON NON-PARAMETRIC OUTLIER DETECTORS

Control chart is a useful device to monitor the stability of any process with an objective to detect unfavorable variations in the parameters (location and scale) of the process. Primarily, control charts are used to monitor manufacturing processes but nowadays they are extensively used in the diagnostic laboratories, health sector, education sector and nuclear engineering. Generally, control chart works into two stages named as prospective stage (Phase I) and retrospective stage (Phase II). Prospective stage is used to analysed the unknown parameters of the process and on the base of these estimated parameters (calculated through Phase I), retrospective stage is used to monitor the quality characteristic of the process. So, consistency of the Phase I analysis may increase the reliability of the Phase II study. Usually, variations are the inherent part of the process. However, outlier detectors are used to detect significant variations or extreme observations in the data. Some detectors are theoretically designed for normally distributed data (Student type and Grubbs type detectors) and others are for non-normal data (i.e. robust Tukey and MAD_e). In this project, we will be proposed non-parametric outlier detectors in the control charting setup for efficient/robust monitoring of process parameters in presence of outlier. We will also examine the performance of the stated proposal in terms of probability to signal, average run length and the comparison of its detection ability with its counterparts. Our study will also cover real life application from Electrical processes (Monitoring the charge of Z-source inverter), adhesive manufacturing industry (monitoring the thickness of tape), and semiconductor manufacturing (monitoring the mass flow controller) to explain the significance of the proposal in different disciplines.