Control charts are graphical techniques for continuous monitoring of the quality of a manufacturing process. Their primary objective is to distinguish between chance and assignable causes of process variation. The chance cause is part of a stable system and is usually very small in magnitude while an assignable cause is due to factors that are not part of the process. When a process operates in the presence of assignable cause, it is not stable and is out-of-control. Control charts can help quickly detect the formation of assignable causes of process disturbances so that investigation and corrective measure is taken before many nonconforming units are produced. In general, control charts are effective tools in eliminating process variability as well as estimating the of the production process(1).
A control chart consists of three horizontal lines, the upper control limit (UCL), the centerline (CL) and the lower control limit (LCL). A process is considered stable, i e, in-control when the plotting points falls within the control limits. A point outside the control limits indicates an out-of-control signal and requires corrective action to bring the process back in-control and improve the quality of the process. The three types of control charts widely used in practice include Shewhart control chart, cumulative sum (CUSUM) chart and exponentially weighted moving average (EWMA) chart.(2)
The performance of these control charts are often compared in terms of their average run length (ARL) properties. ARL represents the average number of samples plotted on a control chart until an out-of-control sample is observed. It measures how quickly a chart responds to process Generally, the ARL for an in-control process should be high, and low, when the process mean shifts to an unsatisfactory level. Statistically, the Shewhart charts are slow in detecting small shifts in the process but handles large shifts perfectly while CUSUM and EWMA charts are very good with small shifts, [3] and [4]
Several authors have studied the ARL performance of these control charts, but most of the reports in the literature are based on simple random sampling (SRS) which is considerably less effective in estimating the population mean as compared to ranked set sampling (RSS) with the same subgroup size. This sampling technique has proven to be very effective in situations where measurements of quality characteristics of interest are difficult or expensive, but could readily be ordered by visual inspection or some cheap method not requiring actual measurement (5) and (6).
There are, however, some recent researches that used Rss scheme to improve the efficiency of the control in detecting changes in process characteristics. For example, 17 ranked sampling with equal and unequal allocation to develop Shewhart X charts, [8] used several modifications of ranked sampling,( 9) used double ranked sampling [10] used robust ranked sampling and very recently, [11] and [12] used the scheme to develop combined Shewhart-EWMA and combined Shewhart-CUSUM control charts respectively.
While previous studies has shown the statistical significance of RSS based control charts for mean, no attempt have been made to compare the performance of the three commonly used control charts for the same subgroup size with same pair of shifts using RSS. Therefore, this paper investigates the performance of the Shewhart X, CUSUM and EWMA charts using RSS.
Using Monte Carlo simulation, we compute the ARL values for the RSS based Shewhart X, CUSUM and EWMA control charts. Comparisons among the newly developed control charts are made and in addition, we compare the results with the classical control charts using SRS. We also give a real life example to demonstrate the simplicity of the scheme.