1. Introduction
Regardless of how well designed or maintained, any manufacturing process produces inherent
or natural variability as a cumulative effect of unavoidable causes. A control chart is one among
other recognized statistical process control tools that, in general, is proactive and mainly aimed at
monitoring the process [10]. A control chart is designed to accurately identify natural variability in a
manufacturing process as a result of unassignable (or chance) causes, or non-natural variability as a
result of assignable (or special) causes, which characterizes an out-of-control process.
In these cases, the traditional Shewhart control p chart has been widely used mainly because it is
simple and effective but it is not sensitive in detecting small shifts that are common in today’s precise
manufacturing processes. This difficulty increases the risk of deciding that the process is under control
when it really is out of control (beta risk of non-detection) or that the process is out of control when
it really is under control (alfa risk of false alarm).
Traditionally, the conventional Shewhart control p charts were constructed based on the normal
approximation for the binomial sampling statistic. But these charts are far from adequate for the
situation of low defect level and/or when the sample size is not large enough, mainly due to skewness
in the exact distribution. For small p values, the binomial distribution is highly asymmetric, and as a
result, any attempt to monitor p with symmetric control limits, is subject to excess of false alarm.
A number of alternatives have been proposed to improve the power and sensitivity of control p
charts in high quality process. A good survey can be found in [16]. But, although these proposed charts
can increase the monitoring accuracy, they still lack the desirable accuracy when the true p is very
small and n is not large.
In a previous paper [9], the authors presented an improved p chart which can provide a considerable
improvement over the usual p chart for attributes. This new chart, based on the Cornish–Fisher
expansion for quantile correction is also better than the traditional Shewhart control chart especially
in the sense that it allows monitoring lower values of p as is the case in high-quality processes. The
proposed method consists in making an adjustment on the control limits that depends only on the
sample size and the value of the process parameters.
This improved control chart can detect large increases in the nonconforming rate p but is not
efficient for detecting small increments of the process parameters. For these situations, an alternative
to control charts for attributes with simple sampling is the application of control charts with double
sampling (DS). Double sampling is a special case of multiple sampling consisting in taking decisions
in two steps instead of in a single step as is usual in control charts.
The DS control chart was firstly proposed by Croasdale [5] in control charts for variables. In this
first DS control chart, information from the first and second samples is evaluated separately, and
confirmation is done only with the second sample. Daudin [6] improved Croasdale’s DS control chart,
and proposed a DS control chart that uses the information from both samples at the second stage. The
larger sample size improves the precision of the control chart since it uses a smaller sample standard
deviation. In estimating the control chart limits, Daudin’s DS control chart is optimized with respect to
the expected sample size. Instead of minimizing the expected sample size, [8] maximized the power
of the control chart to determine the control chart limits. He et al. [7] and Costa and Claro [4] have
made further development of DS control chart for variables.
More recently, research has been conducted to improve the effectiveness of the np charts. Wu
et al. [14] developed an algorithm for the optimization design of the np control chart with curtailment,
considering 100% inspection. Wu and Wang [15] proposed an np chart with a double inspection
feature. The first inspection decides the process state (in control or out of control) according to the
number of non-conforming items found in a sample and in the second inspection, the proposed chart
checks the location of a particular non-conforming item in a sample.
Rodrigues et al. [12] proposed a two-stage sampling or double-sampling (DS) combined with
Shewhart control charts. This is another method used to improve the performance of traditional
Shewhart control p charts, without increasing the (in-control) average number of items inspected per
time unit. During the first stage, one or more items from the sample are inspected and, depending on
the results, the sampling is interrupted or it goes on to the second stage, where the remaining sample
items are inspected.