Proceedings 63rd ISI World Statistics Congress, 11 - 16 July 2021, Virtual P.

000575

Control Charts and Capability Analysis for Statistical Process Control

Arun Kumar Sinha* and Richa Vatsa**
Central University of South Bihar (CUSB, formerly at)
Department of Statistics
Gaya 824236, Bihar
INDIA
**at CUSB
(arunkrsinha@gmail.com & vatsa.richa@gmail.com)
Shewhart initiated the basic work on quality control by making a distinction between chance or
common causes and assignable or special causes of variation. By locating and eliminating
the special causes of variation a process can be brought under statistical quality control. For
this purpose, he developed graphical techniques, known as control charts, based on the
types of available data in manufacturing processes. These charts are also referred to as
Shewhart control charts in modern literature. In this article we discuss the basics of the Shewhart
control charts, their statistical theories and applications with suitable illustrations. These
techniques appear more appropriate for developing countries mainly because of the availability
of natural resources and cheap labor and these could, in turn, help enhance their economies in
short time easily. Besides, we have explained the capability analysis for variable real-life data.
This paper is expected to be of particular interest to those who look for a high-quality
performance in both the manufacturing and the service sectors.
Keywords: Three-sigma limits of control charts, variable control charts, attributes control
charts, specification limits, capability indices
Introduction
Control chart is a statistical tool to detect whether a process is under statistical control or
not. As the appropriateness of a control chart depends on the availability of the data set the types
of available data sets are explained. Illustrations of all control charts are added. Further,
the capability of the process control is explained with illustrations. Besides, the current scenarios
of statistical process control (SPC) in developing countries are discussed because this could
help build their economies quickly. The main advantages of these places are the availability of
natural resources and cheap labor.

2. Control Charts for Statistical Process Control

Though this statistical tool was initially developed to check the current status of the process of
a manufacturing sectors by Shewhart (1924) but with the passage of time this technique has
become
Proceedings 63rd ISI World Statistics Congress, 11 - 16 July 2021, Virtual P. 000576

very popular for examining the process of service sectors too. A process is said to be under
statistical control if all points (statistics) lie within upper and lower control limits. But if a point
exceeds upper control limit (UCL) or falls below the lower control limit (LCL) the process is said
to be out of statistical control. In this situation all points that go outside the control limits are
removed from the given data set and fresh control limits with the remaining observations are
computed. Again, the remaining points are plotted with the new control limits. The process is
continued until all points lie within the control limits. This is referred to as the final control chart
while the first control chart is known as the trial control chart.
3. Types of Data and Control Charts
Usually we come across three types of data in both the manufacturing and service sectors. These
include (i) variable or measurable data, where 𝑥̅ , R, and sigma charts are used (ii) attribute data
(fraction defective or fraction nonconforming) where p chart is employed and (iii) the number of
count of defects or nonconformities (defects) where c chart is found useful. Some authors consider
the attribute data consisting of both fraction nonconforming and nonconformities. See
Montgomery (2009) for more details.
3a. Control Charts for Variable or Measurable Data
For process average we have two types of 𝑥̅ charts, one using range, R and the other involving
standard deviation, σ. For process variability we have two charts, which include R chart and sigma
chart. Let xi j denote the jth measurement of the ith sample, j = 1, 2, . . . ,n and i = 1,2, . . . , k. The
∑𝑛 𝑥
mean of the ith sample is denoted by 𝑥̅𝑖 and it is computed as shown below. 𝑥̅𝑖 = 1𝑛 𝑖𝑗. The range
of the ith sample is represented by Ri and it is obtained by taking the difference between its highest
and the lowest value of the measurements, Ri = Max (xi j) - Min (xi j). Suppose µ denotes the
population mean and σ2 represents the population variance. Then, var(𝑥̅ ) = 𝜎 2 ⁄𝑛 and SE(𝑥̅ )
= σ⁄√𝑛 , where SE stands for Standard Error.

3b. Three-Sigma Control Limits for R and 𝑥̅ Control Charts

UCL= E(𝑥̅ ) + 3 SE(𝑥̅ ) = µ + 3 σ⁄√𝑛, Central Line (CL) = E(𝑥̅ ) = µ, LCL= E(𝑥̅ ) - 3 SE(𝑥̅ ) = µ -
3 σ⁄√𝑛. Note that µ and σ are parameters (standards), which are usually not known. As such these
are, in turn, replaced by their respective estimators. µ is estimated by 𝑥̅̅ ,which is the mean of the
k means. It is computed by 𝑥̅̅ = (𝑥̅1 + 𝑥̅2 + . . . + x̅k)/k, where 𝑥̅𝑖 is the mean of the ith sample. σ
is estimated by R̅/d2 where R̅ =(R1 +R2 + . . . +Rk )/k and d2 is a constant that depends on the sample
size n With these estimators, we obtain the control limits as UCL = 𝑥̿ + 3 R ̅/(d2 √𝑛 ) = 𝑥̿ + A2 R
̅,
CL = 𝑥̿ , LCL = 𝑥̿ - 3 R ̅/(d2 √𝑛 ) = 𝑥̿ - A2 R,̅ where A2 =3/(d2√𝑛). For detecting the lack in the
process variability, usually R and sigma charts are used. Here we discuss R chart. For obtaining
the control limits of this chart we have UCL= D2 σ, CL= d2 σ, LCL= D1 σ. As σ is also estimated
by R̅/d2 we get UCL = D4 R ̅, CL = R̅, LCL = D3 R ̅.The values of constant are available for various
values of n in almost all standard text books of statistics.
Illustrations of R and 𝒙 ̅ Control Charts
To study the aspect of service quality that relates to the amount of time it takes to deliver luggage
(as measured from the time the guest completes check-in proceeds to the time the luggage arrives
in the guest’s room), data were recorded over a 4-week period. Subgroups of five deliveries were
selected from the evening shift on each day for analysis. The data set given below summarizes the
result for all 28 days. The hotel management has instituted a policy that 99% of all luggage
deliveries must be completed in 14 minutes or less.
Day Luggage Delivery Times in Minutes Day Luggage Delivery Times in Minutes
Proceedings 63rd ISI World Statistics Congress, 11 - 16 July 2021, Virtual P. 000577

1 6.7 11.7 9.7 7.5 7.8 15 7.8 9.0 12.2 9.1 11.7
2 7.6 11.4 9.0 8.4 9.2 16 11.1 9.9 8.8 5.5 9.5
3 9.5 8.9 9.9 8.7 10.7 17 9.2 9.7 12.3 8.1 8.5
4 9.8 13.2 6.9 9.3 9.4 18 9.0 8.1 10.2 9.7 8.4
5 11.0 9.9 11.3 11.6 8.5 19 9.9 10.1 8.9 9.6 7.1
6 8.3 8.4 9.7 9.8 7.1 20 10.7 9.8 10.2 8.0 10.2
7 9.4 9.3 8.2 7.1 6.1 21 9.0 10.0 9.6 10.6 9.0
8 11.2 9.8 10.5 9.0 9.7 22 10.7 9.8 9.4 7.0 8.9
9 10.0 10.7 9.0 8.2 11.0 23 10.2 10.5 9.5 12.2 9.1
10 8.6 5.8 8.7 9.5 11.4 24 10.0 11.1 9.5 8.8 9.9
11 10.7 8.6 9.1 10.9 8.6 25 9.6 8.8 11.4 12.2 9.3
12 10.8 8.3 10.6 10.3 10.0 26 8.2 7.9 8.4 9.5 9.2
13 9.5 10.5 7.0 8.6 10.1 27 7.1 11.1 10.8 11.0 10.2
14 12.9 8.9 8.1 9.0 7.6 28 11.1 6.6 12.0 11.5 9.7

[Source: Levine et al. (2016). The data set was collected under Analyze phase of the Measure phase
of Six Sigma project of Beachcomber Hotel.]
We need (i) to draw charts to examine the current status of the hotel service quality.(ii) to examine
whether the process is capable of meeting the 99% goal set forth by the hotel management.(iii) to
calculate the CPU for measuring the process performance.
Solution: (i) First we compute mean and range for each day, and these values are given in Table
1.
Table1. Mean and range of the delivery of luggage times for all 28 days

Day Mean Range Day Mean Range

1 8.68 5.0 15 9.96 4.4
2 9.12 3.8 16 8.96 5.6
3 9.54 2.0 17 9.56 4.2
4 9.72 6.3 18 9.08 2.1
5 10.46 3.1 19 9.12 3.0
6 8.66 2.7 20 9.78 2.7
7 8.02 3.3 21 9.64 1.6
8 10.04 2.2 22 9.16 3.7
9 9.78 2.8 23 10.30 3.1
10 8.80 5.6 24 9.86 2.3
11 9.58 2.3 25 10.26 3.4
12 10.00 2.5 26 8.64 1.6
13 9.14 3.5 27 10.04 4.0
14 9.30 5.3 28 10.18 5.4

The R Chart
We obtain R ̅ =(R1 +R2 + . . . +Rk )/k = 3.482 The control limits for R chart are obtained as UCL =
̅ = 2.114*3.482 = 7.361, CL = 3.482 and LCL= 𝐷3 R
𝐷4 R ̅ = 0.R chart for luggage delivery times
is depicted using Minitab 19 in Figure 1, which does not indicate any range value outside the
control limits or any clear patterns. This means that the process is in statistical control, free from
special cause of variation.
The 𝒙 ̅ Chart
As the R chart shows that the process is in control with respect to the process variability we next
plot the 𝑥̅ chart to examine the current status of the process with respect to the process average.
Proceedings 63rd ISI World Statistics Congress, 11 - 16 July 2021, Virtual P. 000578

For control limits of 𝑥̅ Chart, we have 𝑥̿ =265.38/28 = 9.478, A2 = 0.577, R ̅ =3.482. Finally, we
compute UCL= 9.478 + 0.577*3.482=11.487, CL= 9.478, LCL= 9.478 - 0.577*3.482 =7.469.
Figure 2 displays the 𝑥̅ chart for the luggage delivery times, which does not disclose an average
value of the luggage delivery time outside the control limits. This, in turn, helps conclude that the
process is in statistical control with respect to the process average. As both R chart and 𝑥̅ chart are
in control we conclude that the luggage delivery process is in a state of statistical control.
The rest part of the illustration, (ii) and (iii) are explained under the illustration of Capability
Analysis of Section 4.
Figure 1. R chart for luggage delivery times

3c. Control Charts for Attributes

In manufacturing process attributes of products are referred to as fraction defective or proportion
nonconforming and control charts are called as p-charts. The characteristics are observed by
classifying the products into two classes or groups, usually said as defective or nondetective.
Suppose P denotes the probability of an item being defective while (1-P) shows the probability of
an item being nondetective.

Figure 2. 𝑥̅ chart for the luggage delivery times

𝑑
Let us suppose that the proportion of defective items in then sample of size n is given by p = 𝑛 .
𝑑
It means that d defective items were found out of n items inspected. Then we have E( 𝑛) = P and
𝑑 𝑃(1−𝑃) 𝑃(1−𝑃)
var ( 𝑛)= 𝑛
. Thus, the control limits are given as UCL= P + 3 √ 𝑛
, CL=P, LCL= P - 3
Proceedings 63rd ISI World Statistics Congress, 11 - 16 July 2021, Virtual P. 000579

𝑃(1−𝑃)
√ . As P is usually unknown it is estimated by 𝑝̅ where 𝑝̅ = (d1 + d2 + . . . + dk)/ nk where di
𝑛
defective items were found in the ith sample. Thus, on replacing P by 𝑝̅ we obtain the following
𝑝̅ (1−𝑝̅ ) 𝑝̅ (1−𝑝̅ )
control limits as UCL= 𝑝̅ + 3 √ , CL= 𝑝̅ , LCL= 𝑝̅ - 3 √
𝑛 𝑛
Illustration of p chart
For 32 days, 500 film canisters were sampled and inspected. The given data set lists the number
of defective canisters (the nonconforming items) for each day (the subgroup). Construct p chart
and examine the state of statistical control. {Source Levin at al.(2016).]
Day : 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Items: 26 25 23 24 26 20 21 27 23 25 22 26 25 29 20 19
Day : 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32
Items: 23 19 18 27 28 24 26 23 27 28 24 22 20 25 27 19
Solution
Here n = 500, 𝑝̅ = Total number of nonconforming items/Total items inspected
=761/500*32=761/16000=0.0476. Here UCL = 0.0761, CL=0.0476, LCL=0.0190. See Figure 3.
As all points lie between UCL and LCL we conclude that the process is in a state of statistical
control.
3d. Control Charts for Nonconformities (Defects) or c Chart
The failure to meet a specification at a point results in a defect or nonconformity. An item or a
product having at least one defect or nonconformity is referred to as defective or nonconforming
item. C chart is plotted for defects or nonconformities per unit,. The unit considered may be a
single item or a group of items, part of an item, etc. There are numerous opportunities for defects
to occur on an item but the probability of the occurrence of a defect is very small, For these reasons
the statistical theory for c chart is based on the Poisson distribution. As mean = variance in this
distribution, mean = c and variance = c. Thus, the control limits for nonconformities or c chart
with standard given are given as UCL = c + 3 √𝑐, CL= c, LCL= c - 3 √𝑐.
As c being the parameter of the Poisson distribution is not known, it is estimated by the average
value of the observed values of the nonconformities. Let us represent it by 𝑐̅. The control limits of
c chart with the estimated value as 𝑐̅ we have UCL = 𝑐̅ + 3 √𝑐̅, CL= 𝑐̅, LCL = 𝑐̅ - 3 √𝑐̅.
Figure 3. p chart of number nonconforming

Illustration of c Chart
The following data were collected on the number of nonconformities per unit for 10 time periods.
Proceedings 63rd ISI World Statistics Congress, 11 - 16 July 2021, Virtual P. 000580

Time: 1 2 3 4 5 6 7 8 9 10
Nonconformities per unit: 25 11 10 11 6 15 12 10 9 6
Construct the appropriate control chart and determine UCL, CL and LCL. What is the state of
statistical quality control?
[Source: Levine et al. (2016)]
Solution: We get 𝑐̅ = 11.5 and UCL=21.6735, CL = 11.5, LCL= 1.3265
The control chart is shown in Figure 4. Yes, time 1 observation is above UCL that reveals that the
process is not in statistical control because of the presence of some special causes of variation.
4. Capability Analysis
For analyzing the capability process, we estimate the percentage of products or services that are
within specifications. Assuming that the process is in control and X is approximately normal, we
can estimate the probability of a process outcome being within specifications.
If lower specification limit and upper specification limit are represented by LSL and USL
respectively, then Prob (outcome within specifications) = Prob (LSL< X < USL)
̅/d2 < Z < (USL- 𝑥̿ )/ R
= Prob [(LSL- 𝑥̿ )/ R ̅/d2] where Z= (X- 𝑥̿ )/ R
̅/d2
Illustration of Capability Analysis
See (ii) and (iii) part of the illustration of 𝑥̅ and R charts.It is further mentioned that the hotel
management has instituted a policy that 99% of the luggage deliveries must be completed in 14
minutes or less. Thus USL=14, n = 5, 𝑥̿ = 9.478, R ̅ = 3.482 and d2 = 2.326.
Prob (outcome within only upper specification) = Prob [Z < (USL- 𝑥̿ )/ R ̅/d2]=Prob[Z<(14-
9.478)/(3.482/2.326)] = Prob[Z<3.02] = 0.99874.
This suggests that 99.874% of the luggage deliveries will be made within the specified time. We,
therefore, conclude that the process is capable of meeting the goal of 99% set forth by the hotel
management. (iii) As CPU = (USL – 𝑥̿ ) / 3(R ̅/d2), we get CPU = 1.01. This indicates that the USL
is slightly more than 3 standard deviations above mean.

Figure 4. c chart of nonconformities per unit

5. Scenarios in Developing Countries

SPC techniques played a very important role to help rebuild the economic conditions of Japan after
the second world war under the supervision of W. E. Deming, a US expert of statistical quality
control and improvement. In developing countries SPC techniques can help improve their
economic conditions tremendously if employed properly because of their cheap labor and natural
Proceedings 63rd ISI World Statistics Congress, 11 - 16 July 2021, Virtual P. 000581

resources. Madanhire and Mbohwa (2016) described the applications of SPC in Zimbabwean
manufacturing industries without mentioning a data set. They also discuss the existing associated
problems with manufacturing industries there, which could be minimized. Abtew et al. (2018) used
the data set of garments producing company known as ‘Silver Spark Apparel Limited (SSAL)’, a
Raymond Group Company, India for an illustration. But the used data set is not given in their
paper. There are good garment companies in Ethopia, China, India, etc. and other industries in
developing countries, which could contribute to the incomes of their countries considerably using
the SPC techniques. These countries could follow the example of Japan for achieving their goals
References
Abtew, M. A., Kropi, S., Hong, Y. and PU, L. (2018). Implementation of statistical process control
(SPC) in the sewing section of garment industry for quality improvement. AUTEX
Research Journal, 18 (No. 2), 160 -172.
Levine, David M., Stephen, David F., and Szabat, Kathryn, A.(2016).Statistics for
Managers. Using Microsoft Excel, 7th Ed., Pearson India Education Services Pvt., Ltd,
Noida 201309, UP, INDIA.
Madanhire, I. and Mbohwa, C. (2016). Application of statistical process control in SPC
manufacturing industry in a developing country. Procedia CIRP 40, 580-583 (Available
online at www. sciencedirect.com).
Montgomery, Douglas C. (2009). Statistical Quality Control: A Modern Introduction. 6th ed.,
Wiley, Authorized reprint in 2011 by Wiley India (P) Ltd., New Delhi 110002, INDIA.
Shewhart, W. A. (1924). Some applications of statistical methods to the analysis of physical and
engineering data. Bell Technol. J,, 3, 43-87.