Shift detection and process adjustments for processes with trends

• Main Author: Fahmy, Hesham M
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01-01-2005
• Subjects: Industrial engineering
• ISBN: 9780542411410; 0542411415

One of the key characteristics to be monitored is the mean of the quality characteristic of interest. Even though methods such as design of experiments are used to determine the optimum setting of the process parameters to ensure that the mean of the quality characteristic is within the specified levels, the mean may gradually increase/decrease in linear/nonlinear fashion. This phenomenon is quite related to the change-point detection research that has been investigated by others. In this dissertation, we develop statistical procedures to monitor and adjust such trended processes. We develop a new control chart approach for linear trend detection in the process mean. The approach is described and its performance is compared with cumulative sum (CUSUM), exponentially weighted moving average (EWMA), Shewhart, and generalized likelihood ratio (GLR) charts. The results indicate that proposed approach is effective in detecting small to large trends. We also investigate the run length properties of the proposed approach under linear trends and compare its values with simulation results. Furthermore, we introduce a procedure to detect the drift time of processes subject to linear and nonlinear trend. The procedure is effective in detecting the drift time as early as possible. The confidence intervals for the drift time are estimated and the performance of the proposed estimator is compared with modified CUSUM and EWMA change point estimation procedures. Unlike the traditional statistical process control procedure where the process is investigated once the SPC chart signals that the process is out-of-control but no corrective action is determined, we develop a new adjustment procedure based on the maximum likelihood estimate of the drift time to keep the process on target. We analyze and compare the performance of the adjustment procedure with Single and Double EWMA controllers. Finally, we investigate the integration of an automatic process controller with a monitoring scheme (control chart) to minimize the average penalty per unit in linearly trended processes.