SPC charting procedure for monitoring of small and large shifts in process mean

• Main Author: Masood, Ibrahim
• Format: Thesis
• Language: English
• Published: 2004
• Subjects: TS155-194 Production management. Operations management
• Online Access: http://eprints.uthm.edu.my/7652/1/24p%20IBRAHIM%20MASOOD.pdf; http://eprints.uthm.edu.my/7652/

The research obj ectives are to study the effectiveness of traditional control charts that are Shewharf, Two-Sided CUS1ll11 and EWlvfA in monitoring small and large process mean shifts and to propose an improved statistical process control charting procedures that effective for monitoring all process mean shifts. Process mean shift can be described as unstable patterns such as shift pattern itself and trend pattern. Average run length (ARL), Type I Error and Type II Error are used as the performance measures. The charting procedures were coded in MATLAB program and ex1:ensive simulation experiments were conducted. Design of Experiment (DOE) methods were applied in selecting the suitable design parameters of control charts before conducting the detail ARL simulations. The ARL simulation identifies each control chart monitoring advantages and disadvantages. In general, Two-Sided CUS1l111 and EWlvfA were confirmed effective for detecting small process shifts, while Shell'harf only effective for large process shifts. Specifically, Two-Sided Cusum with (k, h) = (0.5,4.77) and (0.75, 3.34) were identified produced small Type I error, so effective for monitoring small process mean shift, more effective than ETYMA and close to Nelson's Run Rules performance for 0.75cr to 2.5cr shift range. The findings were validated using a few real process data. The concurrent application of Shewharf, Two-Sided CUS1ll11 with (k, h) = (0.5,4.77) and (0.75, 3.34) were proposed as an improved charting scheme. It is observed more effective than the Combined Shewharf-CUS1ll11 which was recommended by Lucas (1982). However, Nelson's Run Rules is not recommended because it provides large Type I error even so effective for monitoring process shift. The findings were confirmed and detailed the individual Nelson Run Rules performances stated by Nelson (1985), except rules for detecting stratification. Finding on different rules gave different 'rate of false signal' (RFS), contradicted with result based on Monte Carlo method (Nelson, 1985) but confirmed the result from Trietsch (1997). Findings on EWMA were confirmed Montgomery (1996) which stated that small constant (A) more sensitive for identifying small shifts while large A better for identifying large shifts.