Control Charts for Variables in Quality Control - Part 1

Dr. Saeed Tasouji Hassanpour

1 Introduction
Control charts are a fundamental tool in the field of quality control. They are graphical representa-
tions of process data over time and are used to monitor and analyze the stability and performance
of a process. Created by Walter A. Shewhart in the 1920s, control charts help in identifying vari-
ation within a process, distinguishing between common cause variation (inherent to the process)
and special cause variation (due to external factors).

ˆ Common Cause Variation: This type of variation is always present in a process, arising
from inherent factors such as:

- Slight wear and tear on machinery over time (machines might experience wear and tear
as moving parts grind against each other, leading to eventual degradation of components
or tools may become dull after repeated use).
- Minor fluctuations in power supply affecting production speed.
- Normal variation in the skills or techniques of operators performing a task.
- Slight differences in raw materials, like small variations in the chemical composition of
metal used in manufacturing.
- Environmental conditions such as regular changes in humidity or room temperature,
which may slightly affect production.

ˆ Special Cause Variation: These are variations due to identifiable, external factors. Exam-
ples include:

- A sudden power outage during production that halts operations.

- A machine breaking down, causing irregular production output.
- Using a batch of defective raw materials, leading to out-of-specification products.
- An operator error, such as incorrect setup of a machine for a specific production run.

1
 - Introducing a new tool or technique that has not been fully integrated into the process,
resulting in unexpected defects.

As shown in Figure 1, these charts display process data points plotted in a time-ordered sequence,
often with a central line representing the mean or average value, along with upper and lower control
limits, usually set at three standard deviations from the mean. The key components of control charts
are:

1. Central Line (or Mean Line): This represents the average value of the process data.

2. Control Limits: Upper and lower control limits indicate the boundaries beyond which the
process output is considered statistically unusual.

ˆ Upper Control Limit (UCL): Usually set at three standard deviations above the
mean, indicating the point beyond which the process output is considered abnormal.
ˆ Lower Control Limit (LCL): Set at three standard deviations below the mean.

Quality Characteristic

UCL

CL

LCL
Sample Number

Figure 1: Quality Control Chart

Control charts serve several purposes in quality control:

ˆ Monitoring Variation: They help identify whether a process is stable or if variations are
occurring, indicating a need for investigation.

ˆ Detecting Outliers: Control charts assist in quickly identifying outliers or unusual patterns
in the data that may indicate a specific problem in the process.

ˆ Determining Process Capability: They provide insights into the process’s ability to meet
specified requirements and identify if the process is in control and capable of producing within
defined limits.

2
 ˆ Supporting Decision Making: By distinguishing between common cause and special cause
variations, they guide decision-making regarding process adjustments or improvements.

Common types of control charts include:

ˆ X-bar (X̄)-chart: Used to monitor the average or central tendency of a process. For example,
if a factory produces screws, the X-bar chart can track the mean diameter of screws produced
over time.

ˆ R-chart (Range chart): Monitors the variability within a process. It is useful when tracking
the consistency of measurements, like the variation in the diameter of screws.

ˆ P-chart (Proportion chart): Used for attribute data, such as the proportion of defective
items in a batch. For instance, a P-chart could monitor the percentage of faulty screws in a
daily production run.

ˆ C-chart (Count of defects chart): Used to monitor the number of defects when the sample
size varies. It is useful in scenarios like tracking the number of errors in a batch of software
tests.

Using three standard deviations for control limits in a control chart helps capture almost all the
normal variation in a process. In a normal distribution (bell curve), most data points fall close to the
average or mean. About 99.7% of the data will be within three standard deviations from the mean.
By setting the control limits at three standard deviations, we can tell if the process is behaving
normally (within expected variation) or if something unusual is happening (when data points fall
outside these limits). In other words, three standard deviations allow us to detect when there is a
rare or unusual event in the process, signaling potential problems or special cause variations that
need attention.
Utilizing control charts in quality control enables organizations to maintain consistency and
reliability in their processes. By systematically monitoring and interpreting data patterns, they
can take proactive steps to ensure product or service quality and identify areas for continuous
improvement. Control charts provide a visual method for observing the behavior of a process over
time, making it easier to detect unusual variations that could indicate potential problems.
Control charts are broadly categorized into two types:

ˆ Control Charts for Variables: These monitor measurable characteristics, such as the
diameter of a screw, the weight of a product, or the temperature during a production process.

ˆ Control Charts for Attributes: These monitor non-measurable characteristics, such as

whether a product is defective or non-defective. Attribute charts focus on countable data,
like the number of defects in a batch.

3
 This lecture primarily focuses on the construction and interpretation of control charts for vari-
ables, as they are commonly used in manufacturing and other processes involving continuous data.
It introduces techniques for crafting control charts and explains specific charts like:

ˆ X-chart: This is used for monitoring the process mean. It helps identify shifts in the central
tendency of a process, such as changes in the average diameter of screws being produced.

ˆ R-chart and S-chart: These charts are used to track process variability. The R-chart
monitors the range within samples (the difference between the smallest and largest values),
while the S-chart focuses on the standard deviation within samples, which is more informative
for larger sample sizes.

Example: Consider a manufacturing unit producing screws. The diameter of each screw can
be measured, making it a variable characteristic. The X-chart is used for monitoring the mean
diameter over time, ensuring that the process stays consistent and produces screws with the correct
dimensions. Meanwhile, the R-chart is used to observe the variability in diameters produced,
ensuring that the difference between the smallest and largest diameters within a sample remains
within acceptable limits. By tracking both the mean and variability, the manufacturer can maintain
high-quality output.
This lecture also introduces Process Capability Analysis, a method used to evaluate if a process
can meet predefined specifications. Process capability analysis is important because even if a process
is in control, it may not necessarily produce items that meet customer or industry specifications. A
common measure in this analysis is the process capability index (Cp), which compares the spread
of the process variability to the allowed specification limits.
Example: A soft drink bottling company might use process capability analysis to ensure that
the filling process consistently maintains the specified volume in each bottle. If the target volume
is 500 mL, with an upper specification limit (USL) of 505 mL and a lower specification limit (LSL)
of 495 mL, the company would measure the capability of the filling machine to consistently fill
bottles within this range. A high process capability index (Cp) would indicate that the process is
well-controlled and that the bottles are being filled accurately.

Objectives

Objectives of this unit include:

ˆ Explaining different types of control charts for variables, such as those used in monitoring the
diameter of screws in manufacturing.

ˆ Determining the appropriate control chart for a given situation, like choosing between an
X-chart and an R-chart based on the type of data and quality characteristic.

4
 ˆ Describing the process for constructing a control chart, such as plotting the mean diameters
of screws over time on an X-chart.

ˆ Constructing and interpreting the control chart for process mean (X-chart), for instance,
plotting the average diameter of screws over time to detect any shifts or trends.

ˆ Constructing and interpreting control charts for process variability (R-chart and S-chart), like
observing the range of screw diameters produced to ensure consistency.

ˆ Defining concepts like specification limits, natural tolerance limits, and process capability, as
applied in ensuring that screw diameters remain within acceptable bounds for quality.

By the end of this lecture, learners are expected to understand various types of control charts for
variables, their applications in different scenarios (such as manufacturing processes), the methods
for construction and interpretation, and critical concepts related to process capability and tolerance
limits. The knowledge gained will allow students to select the appropriate control chart for any
given process and interpret its results to drive quality improvements effectively.

1.1 Steps for Drawing Control Charts for Variables

1. Select the Quality Characteristic: In any manufacturing environment, a product com-
prises several distinct quality characteristics that collectively contribute to its overall qual-
ity and performance. These characteristics can range from tangible physical attributes to
performance-related parameters. Effective quality control necessitates the careful selection of
a specific quality characteristic to monitor closely through a control chart.
For instance, in the manufacturing of screws, various attributes such as length, diameter,
coating, and tensile strength are vital determinants of a screw’s quality. Each of these at-
tributes may play a significant role in the overall performance and reliability of the screws.
However, it’s impractical to simultaneously monitor and address every single characteristic
due to resource constraints.
This is where the Pareto diagram, developed by Joseph Moses Juran and based on the Pareto
principle (or the 80/20 rule), emerges as an invaluable tool. The Pareto diagram aids in prior-
itizing which quality characteristics to focus on by showing the frequency of defects associated
with each specific characteristic. By analyzing this graphical representation, manufacturers
can effectively identify and focus on the most critical areas requiring quality control attention.
By utilizing the Pareto diagram, manufacturing entities can determine and allocate resources
to those particular characteristics that contribute most significantly to defects or quality
issues. This focused approach enables targeted efforts to enhance product quality and ad-
dress deficiencies in the manufacturing process. Through Pareto analysis, manufacturers can

5
 Figure 2: Quality characteristics for screw

strategically direct their quality control measures toward the most influential factors affecting
product quality.
Understanding the Pareto Diagram:

ˆ The X-axis of the Pareto diagram represents the individual quality characteristics or
parameters under consideration. In the context of screw manufacturing, this axis might
display attributes such as ‘Length,’ ‘Diameter,’ ‘Coating,’ and ‘Tensile Strength.’
ˆ The Y-axis displays the percentage or frequency of defects associated with each respec-
tive quality characteristic. This axis quantifies the degree of issues observed for each
parameter, like the proportion of defective screws linked to a specific attribute.

Example: If the ’Diameter’ attribute exhibits the highest bar in the Pareto diagram, it
indicates that a significant proportion of defective screws are due to issues related to their
diameter. Meanwhile, if ’Coating’ or ’Tensile Strength’ has shorter bars, it implies that defects
related to these characteristics are less frequent in comparison.
Table 1 presents quality characteristics data obtained from a sample of 500 screws, among
which 90 products were defective. It illustrates the distribution of defects across various
quality characteristics observed in the sample. Each row corresponds to a specific quality
characteristic, displaying the serial number, the characteristic itself, the frequency of defects
observed, and the percentage of defects it represents out of the total observed defective prod-
ucts.
Analysis: The table details the distribution of defects, highlighting that out of the 500
screws sampled, 90 products were defective. The defects were distributed across various
quality characteristics, with each category representing a proportionate percentage of the total

6
Table 1: Quality Characteristics Data for a Sample of 500 screws with 90 defective products

Serial No. Quality Characteristic Frequency of Defects % of Defects

1 Length 25 27.78
2 Diameter 30 33.33
3 Coating 12 13.33
4 Tensile Strength 8 8.89
5 Others 15 16.67
Total 90 100

observed defective products. For instance, for the Length factor (Serial No. 1), 25 defects
were identified, accounting for 27.78% of the total observed defective products. The final row
summarizes the total counts, presenting the overall sum of defects observed across all quality
characteristics. The total of 90 defects accounts for 18% of the entire sample, showing the
representation of observed defects in the sampled screws. It’s noteworthy that the “Others”
category represents defects or issues that don’t fall within the explicitly categorized quality
characteristics. For example, surface finish, packaging defects, and material imperfections
may be included here.

33.3

30 27.8
% of Defects

20
16.7
13.3

10 8.9

Diameter Length Others Coating Tensile Strength

Quality Characteristics

Figure 3: Pareto Diagram for Quality Characteristics

From the Pareto diagram in Figure 3, it’s clear that diameter and length significantly con-
tribute to non-conforming screws. Thus, these characteristics should be prioritized for quality
control.

7
1.2 Selecting the Type of Control Chart
The choice of a control chart is crucial in quality control and is influenced by various factors,
including the nature of the measurement, the specific quality characteristic, and associated cost
implications.

1.2.1 Determining Chart Category: Variables or Attributes

The selection of a control chart begins by categorizing the nature of the quality characteristic
to be controlled. If the characteristic is measurable—such as weight, length, diameter, or other
quantifiable parameters—the appropriate choice is the control chart for variables. For instance, in
the context of production, where measurable quality characteristics like weight and diameter are
chosen for control, the logical decision is to utilize control charts for variables to monitor these
aspects.
Example: Consider a factory that produces metal rods, and one of the critical quality charac-
teristics is the diameter of these rods. Since diameter is a quantifiable measurement, the factory
would use a control chart for variables to monitor any variations in the rod diameter.
Conversely, when dealing with non-measurable characteristics such as color, surface roughness,
or other qualitative traits, the suitable choice is the control chart for attributes. These types of
charts help in monitoring attributes that cannot be precisely measured but can be categorized or
counted based on certain criteria.

(a) Measuring the diameter of a metal rod. (b) Monitoring color consistency.

Figure 4: Quality control processes: metal rods (left) and painted parts (right).

Example: In a factory producing painted parts, the quality control team may monitor the
color consistency or surface finish using control charts for attributes, as these characteristics are

8
qualitative and not easily measurable.

1.2.2 Choosing the Specific Chart

Once the category (variables or attributes) has been identified, the next step involves selecting the
specific chart to be used. This choice depends on understanding the particular situation and the
main goal of the control. It requires knowledge about the available charts, their applicability to the
given scenario, and aligning the chart’s objectives with the objectives of the control process.
The process of selecting a control chart involves a systematic assessment of the characteris-
tic, its measurability, cost implications, and the specific purpose of the control. It is a critical
decision-making process that ensures effective quality control and resource optimization within the
manufacturing or quality management framework.

1.3 Selection of Rational Subgroups

Control charts for quality characteristics are constructed using sample data, raising the question
of how data are gathered for analysis and process control. To address this, Walter A. Shewhart
introduced the concept of rational subgroups, a fundamental principle in the realm of quality control.
Shewhart’s insight was to categorize data into rational subgroups to distinguish between types of
variations observed in a process. He proposed that differences between rational subgroups serve as
indicators of process changes, often attributed to specific assignable causes. In contrast, differences
observed within a rational subgroup signify inherent variability—termed as chance causes. Hence,
the methodology aims to separate data in such a way that assignable causes, if present, exhibit as
differences between subgroups rather than within them.
A rational subgroup represents a small, specifically chosen set of items or units that are produced
under similar conditions within a relatively short time frame. The intention is to minimize external
influences and ensure that the variation within the subgroup primarily stems from chance causes
rather than systematic process changes.
In industrial contexts, the term ’sample’ is frequently used interchangeably with ’subgroup.’
Thus, within the framework of this course, both terms, ’sample’ and ’subgroup,’ are employed
synonymously.

1.3.1 Practical Implementation and Examples

Example 1: Imagine a company that produces bottles of a specific size and shape. The quality
control team is responsible for ensuring that the bottles meet a certain volume requirement. They
use a control chart to monitor the volume of the bottles produced. The production process involves
molding the bottles using a machine. To apply Shewhart’s concept, the team selects a specific

9
number of bottles produced consecutively within a short timeframe, under similar conditions, for
measurement. This set of, let’s say, 20 bottles, constitutes a rational subgroup or a sample.
The team measures the volume of these 20 bottles and records the data. The idea behind
choosing these 20 bottles as a rational subgroup is to capture the inherent variability within a short
time frame and similar production conditions. Any differences in the volumes within this subgroup
are considered as resulting from chance causes inherent in the process, such as minor fluctuations
in material density or slight variations in ambient temperature.
Now, let’s suppose the next set of 20 bottles produced, shortly after the first subgroup, shows
a significant change in average volume compared to the first subgroup. This difference between
the averages of these two rational subgroups could indicate an assignable cause, such as a machine
calibration issue, change in the raw material, or some other factor affecting the production process.
By observing differences between rational subgroups, the quality control team can quickly iden-
tify when there might be a specific assignable cause impacting the process. This helps them take
corrective actions promptly to maintain the quality of the bottles.
Example 2: For instance, in a manufacturing setting producing screws, a rational subgroup
might comprise screws manufactured using the same machine settings, material batch, and within
a specific timeframe. By ensuring consistency in production conditions, any variations observed
between such rational subgroups could likely signal external or assignable causes affecting the
manufacturing process. Conversely, variations within these subgroups may predominantly represent
the inherent variability expected due to random factors.
The deliberate selection of rational subgroups facilitates the differentiation between variations
attributable to chance causes and those arising from identifiable process changes. This separation
aids in effectively identifying and addressing assignable causes, contributing to the optimization of
the quality control process.

1.4 Size of Subgroup (Sample Size)

In the context of quality control, determining the size of the subgroup or sample is crucial to ensure
maximum uniformity within the dataset. Ideally, the sample should be kept as small as possible
to maintain consistency. However, in industrial practice, sample sizes typically range between four
and five units or items.
From a statistical perspective, a sample size of four or more units holds significance. According
to sampling distribution theory, even if the population distribution is non-normal, the distribution
of the sample mean X tends to approximate normality for samples consisting of at least four items.
This understanding is particularly advantageous when interpreting control chart limits, as it ensures
a more reliable representation of the process’s behavior.
When aiming to increase the sensitivity of the control chart, larger sample sizes—such as 10

10
or 20 units—are utilized. This choice is founded on the principle that the standard error of a
statistic is inversely proportional to the sample size. As the sample size increases, the standard
error decreases. This is due to the nature of the standard error formula, which is the standard
deviation of the sampling distribution. The standard error is inversely proportional to the square
root of the sample size. Consequently, the 3σ limits (upper and lower control limits) lie closer to
the center line, enhancing the chart’s sensitivity to detect process variations.
However, in scenarios where the manufactured items are either expensive or are destroyed during
inspection, smaller sample sizes of 2 or 3 are adopted. This choice is a trade-off between sensitivity
and resource efficiency, where the limitations in sample size are compensated to manage cost or
resource constraints.

1.4.1 Practical Application and Examples

For instance, in a production facility manufacturing electronic components, a sample size of 5

might be selected for routine quality checks due to its ease of computation and relatively reliable
representation of the production process. On the other hand, if the manufacturing process undergoes
modifications or improvements, using larger sample sizes of 10 or 20 might offer a more sensitive
depiction of the process changes, ensuring more accurate identification of variations.
Similarly, in high-cost manufacturing scenarios, such as aerospace parts production or specialized
medical devices, smaller sample sizes of 2 or 3 might be employed to minimize waste while still
offering reasonable insights into the quality of the produced items.
The choice of the sample size plays a crucial role in balancing accuracy, resource utilization,
and the sensitivity of the control chart in reflecting process variations within a manufacturing
environment.

1.5 Frequency of Subgroups (Number of Samples)

Determining the number of subgroups or samples is a critical consideration in quality control. While
there are no rigid guidelines, the selection should be adequate to discern trends, patterns of level
shifts, and process changes. A level shift in quality control refers to a significant and abrupt change
in the average of the process output or measurement. The decision-making process involves finding
a balance between acquiring enough data for effective analysis and managing the associated costs
of sample collection.
Two primary approaches dictate the frequency of sampling:
i) Larger samples at less frequent intervals, or
ii) Smaller samples at more frequent intervals.
It is advantageous to initiate the process by taking samples more frequently. This allows for
an in-depth understanding of the process behavior and aids in swiftly detecting and rectifying

11
variations, particularly at the outset of the production cycle. As the process stabilizes and is
brought under statistical control, the frequency of sampling can be adjusted accordingly.

1.5.1 Practical Implementation and Examples

In an automobile assembly line, where various components are manufactured and assembled, a qual-
ity control team might conduct frequent hourly checks on the initial output. As the manufacturing
processes stabilize and the production lines attain consistent quality, the frequency of checks might
be adjusted to occur every 2-3 hours, promoting efficiency without compromising oversight.
On the other hand, in the event of unexpected quality deviations or process instabilities, the
team may increase the sampling frequency to swiftly identify the root causes of the issues. This
adaptive sampling strategy allows for prompt corrective actions, ensuring the overall quality of the
manufactured components.
The decision on the frequency of subgroups or samples plays a vital role in maintaining process
control and ensuring product quality within a manufacturing environment.

1.6 Designing Data Collection Forms

Once decisions are made regarding the quality characteristic, subgroup/sample size, and the number
of subgroups, the next step involves the design of a data recording form. The form’s structure and
content are crucial and should align with the specific control chart that will be used for analysis.
For instance, let’s consider a typical data recording form for the X-chart:

Data Recording Form for X-Chart

The data recording form typically includes two parts. The top segment of the form contains essential
information related to the part name, operation machine, gauge used, unit of measurement, and
specifications. This top section ensures a comprehensive understanding of the context surrounding
the collected data.
The lower segment of the form is devoted to the actual data collection, providing spaces for sub-
group/sample numbers, the date and time of sample collection, and the raw observations obtained.
The ”Measurements” section has sub-columns categorized into X1 to X4, representing different
dimensions or characteristics. Additionally, mean values and standard deviation (SD) provide an
overview of the measurements’ central tendencies and their variations.
Interpretation of the Table:
The table represents a sample data recording form for the X-chart used in quality control. Each
row represents a subgroup of items or units produced under similar conditions within a specific time
frame. The columns contain essential information such as subgroup number, date, time of sampling,
and four measurements denoted as X1, X2, X3, and X4, corresponding to distinct characteristics or

12
 Table 2: Updated Data Recording Form for X-Chart with Bessel’s Correction

Subgroup Date Time Measurements Mean SD

X1 X2 X3 X4
1 10/01 09:00 5.6 5.5 5.7 5.6 5.60 0.08
2 10/01 10:30 5.7 5.8 5.7 5.6 5.70 0.08
3 10/01 12:00 5.6 5.7 5.7 5.8 5.70 0.08
4 10/01 13:30 5.9 5.8 5.7 5.9 5.82 0.10
5 10/01 15:00 5.6 5.7 5.6 5.7 5.65 0.06
6 10/01 16:30 5.8 5.9 5.8 5.7 5.80 0.08
7 10/01 18:00 5.6 5.7 5.8 5.9 5.75 0.11
8 10/01 19:30 5.7 5.8 5.8 5.7 5.75 0.08
9 10/01 21:00 5.8 5.9 5.8 5.9 5.85 0.06
11 10/02 10:30 5.7 5.8 5.7 5.6 5.70 0.08
12 10/02 12:00 5.8 5.7 5.8 5.7 5.75 0.06
13 10/02 13:30 5.7 5.8 5.9 5.9 5.83 0.09
14 10/02 15:00 5.8 5.9 5.8 5.7 5.80 0.08
15 10/02 16:30 5.6 5.7 5.6 5.7 5.65 0.08

dimensions of the product being monitored. The ”Mean” column displays the average value across
the measurements for each subgroup, while the ”SD” column represents the standard deviation,
indicating the variability within the subgroup.
The general formula for the mean of a set of n measurements X1 , X2 , . . . , Xn is given by:

X1 + X2 + · · · + Xn
Mean =
n
For example, in the case of the subgroup with measurements X1 = 5.6, X2 = 5.5, X3 = 5.7,
and X4 = 5.6, the mean is calculated as:

5.6 + 5.5 + 5.7 + 5.6

Mean = = 5.60
4
The formula for the sample standard deviation (SD) of a set of n measurements X1 , X2 , . . . , Xn
with mean X̄ is:
s
(X1 − X̄)2 + (X2 − X̄)2 + · · · + (Xn − X̄)2
SD =
n−1

For the same subgroup with X1 = 5.6, X2 = 5.5, X3 = 5.7, X4 = 5.6, and X̄ = 5.60, the SD is
calculated as:
r
(5.6 − 5.60)2 + (5.5 − 5.60)2 + (5.7 − 5.60)2 + (5.6 − 5.60)2
SD = = 0.08
4−1

13
 This form facilitates systematic data collection and analysis. It records specific measurements
at different time points to observe the variation and central tendencies within the manufacturing
process. Through the mean and standard deviation calculations, this form aids in detecting po-
tential variations or irregularities within the process and assists in making informed decisions for
quality improvement and control.

1.7 Determination of Trial Centre Line and Control Limits

After collecting data and understanding the distribution of the quality characteristics, the next
critical step is determining the trial centre line and control limits for the control chart. These limits
help assess whether the process is within acceptable boundaries or if corrective actions are needed.

1.7.1 Understanding Probability Distributions

Quality characteristics often follow certain probability distributions. Many of these distributions
resemble the normal distribution or can be closely approximated by it. In probability theory, a
normally distributed random variable X will fall within the range of µ−3σ and µ+3σ approximately
99.73% of the time, where:

- µ is the mean of the distribution,

- σ is the standard deviation.

Mathematically, this can be expressed as:

P (µ − 3σ ≤ X ≤ µ + 3σ) = 0.9973 (1)

This implies that the probability of a random variable X falling outside these limits is (1 −
0.9973 = 0.0027), meaning there is a very small chance of this occurring.
Example: If a factory is inspecting 100 items, the probability suggests that approximately 0.27
items might fall outside the µ ± 3σ range. This would indicate that these few items are statistically
unusual and may require further investigation to determine if there is a process issue.

1.7.2 Setting Control Limits

In control charts, when an observation surpasses the 3σ limits, it indicates a significant devia-
tion from the expected process performance, often triggering an investigation or corrective action.
Therefore, control limits are commonly set based on these 3σ boundaries.
For a sample statistic M (such as the sample mean, range, or proportion of defectives), with:

- µM as the mean of the sample statistic,

14
 - σM as the standard error (standard deviation) of the sample statistic M ,

The trial centre line and control limits are determined using the following equations:

Centre line (CL) = E(M ) = µM (2)

Upper control limit (UCL) = µM + 3σM (3)
Lower control limit (LCL) = µM − 3σM (4)

Example: Suppose the mean of a process is 50 units, and the standard error σM is 1. The
control limits for this process would be calculated as:

- Centre Line (CL) = 50,

- Upper Control Limit (UCL) = 50 + 3(1) = 53,

- Lower Control Limit (LCL) = 50 - 3(1) = 47.

If a sample statistic (like the average of a subgroup) falls outside the range of 47 to 53, it signals
that the process may be out of control and requires attention.
These calculated control limits provide a reference for monitoring and maintaining quality within
a process. When observations stay within these limits, the process is considered stable. If not, it
may indicate issues such as assignable causes of variation that need to be addressed.

1.8 Constructing a Control Chart

In the process of constructing a control chart, the determination of the centre line and control limits
plays a critical role. It is important to note that these values differ for each type of chart used, each
being tailored to specific statistical measures. Let’s explore the methodology of establishing these
limits while illustrating each type of chart individually.
Following the calculation of the centre line and control limits based on the gathered data, the
next step involves the actual construction of the control chart. A control chart typically features a
statistical measure (such as mean, range, or number of defects) represented along the Y-axis, while
the sample number or time is displayed along the X-axis.
To visually depict the established limits and mean, the centre line is usually denoted by a solid
line, and the control limits are depicted by dotted lines on the chart. The process of constructing
the control chart involves plotting the value of the statistical measure for each sample against the
corresponding sample number or time.
Consecutive sample points are linked by line segments, facilitating a visual representation of
the data’s trend and variability within the defined control limits. This graphical representation is

15
instrumental in monitoring the process, enabling the easy identification of potential deviations and
facilitating prompt corrective actions when necessary.

Figure 5: Control Chart Example showing the center line, UCL, and LCL.

1.9 Drawing Preliminary Conclusions from the Control Chart

After constructing the control chart, the next essential step is to draw preliminary conclusions by
analyzing the chart. This involves scrutinizing the plotted sample points to observe whether they
fall within or outside the established upper and lower control limits. Any sample points that lie
outside these limits are specifically highlighted by marking them with a dotted circle. It is important
to understand that the control limits on the initial chart are referred to as trial control limits.
What does being ”in control” mean? If all sample points lie within the control limits
and no unusual patterns or trends are observed, this indicates that the process is under statistical
control. In this case, the process is considered stable and operating under the influence of only
chance causes, meaning no significant external factors (known as assignable causes) are impacting
the process.
Example: Consider a manufacturing process where a machine is cutting metal rods. If all the
measurements (e.g., length) fall within the control limits, and there are no recurring patterns like
a consistent upward or downward trend, the process is considered ”in control,” and only random
variations (like minor fluctuations in material quality) affect the output.

1.9.1 Understanding Minor Variations

Even when a process is under statistical control, minor variations may still exist due to factors such
as:

- Machinery wear and tear,

16
Figure 6: Control Chart Example showing Upward, Downward, and Cyclical Trends

17
 - Operator differences in performance,

- Slight variations in raw materials.

These small variations are considered inherent to a stable process and are known as chance
causes. Once confirmed that the process is in statistical control, the trial control limits are accepted
as standard limits, and the values of X, σ, R, etc., become standard benchmarks for monitoring
future production.
Adjusting Control Limits Over Time As more data is collected, there might be a need to
revise the control limits periodically. Regular sampling ensures continuous statistical control, with
samples being taken at specified intervals (e.g., weekly, monthly, or after a set number of items are
processed).
Example: If a company produces 10,000 units per month, they might decide to take a sample
every 1,000 units to ensure the process remains in control. As data accumulates, they may adjust
the control limits to reflect long-term trends in the data.

1.9.2 Detecting Non-Statistical Control

In many initial analyses, processes often show signs of being out of control. This could be indicated
by points falling outside the control limits or by irregular patterns, such as long runs of points
either above or below the center line. These situations signal the presence of assignable causes in
the process, which must be investigated and corrected.
Example: If, during the analysis of a plastic molding process, several points lie beyond the
control limits or show a consistent upward trend, it might suggest that the material being used is
inconsistent in quality, possibly due to a new supplier.
Once the causes are identified—such as poor raw material quality or a machine calibration
issue—corrective actions are taken. These might include switching suppliers or recalibrating the
machinery to ensure consistent performance. By removing the out-of-control points and recalcu-
lating the centre line and control limits, we establish revised control limits. This iterative process
of recalculating and adjusting the limits continues until the process stabilizes and is brought back
under statistical control.

1.9.3 Continuous Process Improvement

Studying how to construct control charts and analyzing whether a process is under or out of control
forms the foundation for further mastery of quality control techniques. Control charts for variables
play a critical role in understanding a process’s behavior and ensuring that it stays within the
acceptable limits, providing an effective way to maintain and improve the quality of products over
time.

18
2 Control Charts for Variables
In any manufacturing process, achieving uniformity in produced items regarding specific charac-
teristics is challenging. Inevitably, variations exist in these items due to factors like raw material
quality, workforce skill levels, machinery, or faulty equipment. Such variations have a direct impact
on product quality, and they may arise from either chance causes (natural variability that occurs
due to minor fluctuations, randomness, or inherent unpredictability within the system) or assignable
causes (factors that are not part of the usual, expected behavior of the process). Statistical Process
Control aims to identify and, if possible, eliminate these sources of variation. Control charts for
variables are utilized when the quality characteristic under scrutiny is measurable.
In quality control, the term “variable” refers to a measurable quality characteristic, such as the
diameter of ball bearings, length of refills, weight of cricket balls, and similar attributes. Control
charts designed for measuring these quality characteristics are known as control charts for variables.
Managing a measurable quality characteristic involves controlling both the central tendency (av-
erage) and the dispersion (variability) of the attribute or process. The mean is typically employed
to regulate central tendency, while range or standard deviation is calculated to control variabil-
ity. Consequently, different control charts for variables are implemented to manage the mean and
variability of the process. The most commonly utilized ones include:

i. Control chart for mean

ii. Control chart for variability

The X-chart, known as the control chart for mean, monitors the average quality level or process
mean. It is essential to consider the effect of process variability on product quality. When the
manufacturing process shows considerable variation, creating a wider range of product values, the
overall quality appears lower. Minimizing this variability is crucial to improve product quality.
The following charts are employed to monitor process variability:

i. Range chart (R-chart)

ii. Standard deviation chart (S-chart)

2.1 Control Chart for Mean (X-chart)

To regulate the process mean within any production process, the control chart for mean, commonly
known as the X-chart, is employed. The X-chart aids in monitoring the variance in the mean of
periodically drawn samples from the process. Instead of individual measurements, sample means
are plotted on the control chart as they offer additional information, making the X-chart more
sensitive in detecting changes in the process mean.

19
 Let’s outline the various steps involved in constructing the X-chart:
Steps Involved in the Construction of the X-chart
Step 1: Selection of the measurable quality characteristic for constructing the X-chart.
Step 2: Determination of the subgroup/sample size, typically four or five items in a sample,
and collecting around twenty to twenty-five samples for the X-chart.
Step 3: Random selection of sample units/items from the process, ensuring each unit/item has
an equal chance of selection.
Step 4: Measurement of the quality characteristic for each selected item/unit in each sample,
ensuring precise measurements are taken using standard instruments and by skilled professionals.
Step 5: Calculation of the sample mean for each sample. If X1 , X2 , . . . , Xn are the measure-
ments for a sample of size n, the sample mean is given by:

n
1X
X̄ = Xi (5)
n i=1

where Xi represents the measurement of the quality characteristic of the ith unit.
Step 6: Setting up the control limits for the X-chart involves pivotal statistical parameters to
determine whether the process is under statistical control. The 3-sigma (3σ) control limits play a
significant role in quality control, aiding in assessing the process state. These limits are instrumental
in identifying the presence of assignable causes or variations within the process that may lead to
defects or inconsistencies in the final product.

Centre line = E(X̄) (6)

Lower control limit (LCL) = E(X̄) − 3SE(X̄) (7)
Upper control limit (UCL) = E(X̄) + 3SE(X̄) (8)

where E(X̄) and SE(X̄) are the mean and the standard error of the sampling distribution of
X̄, respectively.
Here, the Centre line signifies the expected mean value, denoted by E(X̄) , and the Lower
Control Limit (LCL) and Upper Control Limit (UCL) are calculated as three standard deviations
(3SE(X̄)) away from the mean. The range between these control limits serves as a measure for
acceptable variation within the process. Any data point outside these boundaries is indicative of
potential issues or variations in the process.
If X has a normal distribution with mean µ and standard deviation σ, then X̄ has a normal
distribution with mean µ and standard deviation √σ .
n

σ
X̄ ∼ N (µ, √ ) given X ∼ N (µ, σ) (9)
n

20
2.2 Understanding the Statistical Distribution of the Sample Mean
The statistical distribution of the sample mean is fundamental in constructing control charts, specif-
ically the X-chart. If X1 , X2 , . . . , Xn represent a sample of size n drawn from a normal population
with a mean µ and variance σ 2 , the sample mean X̄ is also normally distributed, with a mean µ
and a variance σ 2 /n. This is the statistical basis for determining the control limits, which allow us
to evaluate the stability and performance of the production process.
The standard error of X is the square root of its variance. For the sample mean, the standard
error is given by:

p
SE(X) = V ar(X) (10)
r
σ2 σ
q
SE(X̄) = V ar(X̄) = =√ (11)
n n

Example: Imagine a factory where a machine produces metal rods, and their length must be
monitored. The mean length of the rods is 50 mm, with a standard deviation of 2 mm. If we take
a sample of 10 rods (n = 10), the standard error of the mean is:

2
SE(X̄) = √ ≈ 0.632
10
This means the sample mean will vary by about 0.632 mm on average from the true population
mean.

2.3 Constructing Control Limits

The construction of control limits on an X-chart is crucial for monitoring process performance.
These limits help distinguish between normal variations (due to chance) and unusual variations
(assignable causes) that require further investigation.
For an X-chart, the control limits are calculated as:

Centre line (CL) = E(X̄) = µ

3σ
Lower control limit (LCL) = E(X̄) − 3SE(X̄) = µ − √
n
3σ
Upper control limit (UCL) = E(X̄) + 3SE(X̄) = µ + √
n
√
Alternatively, using the constant A = 3/ n, the control limits can be written as:

21
 LCL = µ − Aσ
UCL = µ + Aσ

Example: In the case of the metal rods, if the mean length (µ) is 50 mm, the standard deviation
(σ) is 2 mm, and the sample size (n) is 10, the control limits are calculated as:

3×2
U CL = 50 + √ = 50 + 1.897 ≈ 51.90
10

LCL = 50 − 1.897 ≈ 48.10

Thus, any sample mean outside the range of 48.10 mm to 51.90 mm would indicate that the
process might be out of control.

2.4 Estimating Parameters for Control Limits

Since the true values of µ and σ are often unknown, they must be estimated from the data collected
when the process is assumed to be under statistical control.
¯ ), which is the average of all
The best estimator of the process mean (µ) is the grand mean (X̄
the sample means:

k k n
¯= 1 1X1X
X
X̄ X̄i = Xij (12)
k i=1 k i=1 n j=1

To calculate the grand mean, first compute the mean of each sample, and then take the average
of these sample means.

2.5 Estimating Standard Deviation (σ)

The standard deviation σ can be estimated using either the sample range (R) or the sample standard
deviation (S).

i. Using Sample Range (R):

ˆ Calculate the range (R) for each sample, which is the difference between the largest and
smallest values in the sample.
ˆ Compute the average range (R̄) across all samples.

22
 ˆ Estimate σ using the formula:
R̄
σ≈
d2
where d2 is a constant depending on the sample size.

ii. Using Sample Standard Deviation (S):

ˆ Calculate the sample standard deviation (S) for each sample.

ˆ Compute the average standard deviation (S̄) across all samples.
ˆ Estimate σ using the formula:
S̄
σ≈
c4

2.6 Control Chart Constants

The constants used for estimating σ and constructing control limits, such as A2 , d2 , and c4 , depend
on the sample size and are tabulated as shown in Table 3.

Table 3: Values for Control Chart Constants (A2 , d2 , c4 )

Sample Size (n) Value of A2 Value of d2 Value of c4 Value of A3

2 1.088 1.128 0.7979 2.659
3 1.023 1.693 0.8862 1.954
4 0.729 2.059 0.9213 1.628
5 0.577 2.326 0.9400 1.427
6 0.483 2.534 0.9515 1.287
7 0.419 2.704 0.9594 1.182
8 0.373 2.847 0.9650 1.099
9 0.337 2.970 0.9693 1.032
10 0.308 3.078 0.9727 0.975
11 0.285 3.173 0.9754 0.927
12 0.266 3.258 0.9776 0.886
13 0.249 3.336 0.9794 0.850
14 0.235 3.407 0.9810 0.817
15 0.223 3.472 0.9823 0.789
16 0.212 3.532 0.9835 0.763
17 0.203 3.588 0.9845 0.739
18 0.194 3.640 0.9854 0.718
19 0.187 3.689 0.9862 0.698
20 0.180 3.735 0.9869 0.679

By using these constants, we ensure that the control limits are set correctly based on the sample
size, making it easier to monitor and control the process effectively.

23
2.6.1 Case I: When Standard Deviation is Estimated by the Sample Range R

In this case, the standard deviation σ is estimated using the sample range R. The estimate of σ is
calculated as follows:

R̄
σ̂ = (13)
d2
where:

- R̄ is the mean of all sample ranges,

- d2 is a constant that depends on the sample size.

Calculating the Mean Sample Range (R̄):

The mean of all sample ranges (R̄) is computed as:

k k
1X 1X
R̄ = Ri = (R1 + R2 + . . . + Rk ) (14)
k i=1 k i=1

Here, Ri is the sample range for the i-th sample, and is calculated as:

Ri = Xmax − Xmin (15)

where Xmax and Xmin represent the maximum and minimum measurements of the quality
characteristic within a sample, respectively.
The constant d2 is derived from statistical tables and depends on the sample size. This constant
is crucial for converting the sample range into an estimate of the standard deviation σ.

Control Limits for the X-chart

¯ and σ by
When estimating µ by X̄ R̄
d2 , the control limits for the X-chart are derived as follows:
1. Center Line (CL): The center line (CL) is the estimate of the process mean, given by the
¯:
grand mean X̄

¯
CL = µ̂ = X̄ (16)

2. Lower Control Limit (LCL): The lower control limit (LCL) is calculated using the estimated
mean and standard deviation:

3 ¯ − √3 R̄
LCL = µ̂ − √ σ̂ = X̄ (17)
n n d2
3. Upper Control Limit (UCL): The upper control limit (UCL) is similarly calculated:

24
 3 ¯ + √3 R̄
UCL = µ̂ + √ σ̂ = X̄ (18)
n n d2
In these formulas:

- µ̂ represents the estimated process mean,

- σ̂ denotes the estimated standard deviation,

¯ represents the grand mean (mean of all sample means),
- X̄

- R̄ is the mean of the sample ranges, and

- d2 is a constant related to the sample size.

2.6.2 Using the Constant A2

In practical applications, the constant A2 is often used to simplify the computation of control limits.
The value of A2 is related to the sample size and is used to replace √3 :
n

3
A2 = √ (19)
d2 n
With this constant, the control limits can be expressed in a simplified form:

¯ − A R̄
LCL = X̄ 2
¯
UCL = X̄ + A2 R̄

Example: Suppose the mean of the sample ranges R̄ is 5, the sample size (n) is 5, and the
corresponding d2 value for n = 5 is 2.326 (from control chart constants table). The estimated
standard deviation is:

5
σ̂ = ≈ 2.15
2.326
¯ is 50, the control limits for the X-chart would be:
Then, if the grand mean X̄

3 × 2.15
LCL = 50 − √ ≈ 47.11
5
3 × 2.15
U CL = 50 + √ ≈ 52.89
5
This approach allows for easy monitoring of process stability using control charts, facilitating
early detection of deviations in the production process.

25
Case II: When the Standard Deviation is Estimated by the Sample Standard Deviation

In this case, the standard deviation σ is estimated using the mean of the sample standard deviations.
The estimate of σ is given by:

S̄
σ̂ = (10)
c4
where:

- S̄ is the mean of all sample standard deviations, and

- c4 is a constant that depends on the sample size.

Calculating the Mean Sample Standard Deviation (S̄)

The mean of the sample standard deviations is computed as follows:

k
1X
S̄ = Si (11)
k i=1

where:

- Si is the sample standard deviation for the i-th sample, calculated by:

v
u n
u 1 X
Si = t (Xj − X̄i )2 (12)
n − 1 j=1

In this formula:

- Xj denotes each data point in the sample,

- X̄i is the sample mean for the i-th sample, and

- n is the number of elements in the sample.

Control Limits for the X-chart

¯ and σ is estimated by S̄, the control limits for the
When the process mean µ is estimated by X̄
X-chart are calculated as follows:

1. Center Line (CL)

¯:
The central line is the estimate of the process mean, which is given by the grand mean X̄

¯
Central Line (CL) = µ = X̄ (20)

26
2. Lower Control Limit (LCL)

The lower control limit is calculated by subtracting three standard errors from the process mean:

3 ¯ − √3 S̄ = X̄
¯ − A S̄
Lower Control Limit (LCL) = µ̂ − √ σ̂ = X̄ 3 (21)
n n c4

3. Upper Control Limit (UCL)

The upper control limit is calculated by adding three standard errors to the process mean:

3 ¯ + √3 S̄ = X̄
¯ + A S̄
Upper Control Limit (UCL) = µ̂ + √ σ̂ = X̄ 3 (22)
n n c4

Constant A3

The constant A3 is used to simplify the control limit formulas. It is defined as:

3
A3 = √
c4 n
This constant depends on the sample size and the c4 constant, which is derived from statistical
tables.

Example

Consider a process where the mean sample standard deviation S̄ is 1.5, the sample size (n) is 5,
and the c4 value for n = 5 is 0.9400. The estimated standard deviation is calculated as:

1.5
σ̂ = ≈ 1.60
0.9400
¯ is 50, the control limits for the X-chart are:
Assuming the grand mean X̄

3 × 1.60
LCL = 50 − √ ≈ 47.85
5
3 × 1.60
U CL = 50 + √ ≈ 52.15
5
These control limits provide a clear boundary for monitoring the process. If a sample mean
falls outside this range, it indicates that the process may be out of control, prompting further
investigation.
Step 7: Construct the X̄-chart
Once the central line (CL) and control limits are established, the construction of the X̄-chart
commences. Plotting the sample mean (X̄) against the sample number, the horizontal axis rep-

27
resents the sample number, while the vertical axis denotes the sample means. These points are
plotted, and successive sample means are connected using line segments.

Step 8: Interpret the Result

Upon completing the chart, the next step is interpreting whether the process is under statistical
control. This involves assessing if all sample points lie within the upper and lower control limits:

- If all points are within the control limits, the process is considered to be under statistical
control, meaning that only random or chance causes are influencing the process.

- If one or more points fall outside the control limits, it indicates that the process is out of
control. This signifies the presence of assignable causes that are affecting the process.

Action if Process is Out of Control: If the process is out of control, the next step is to:

- Investigate and identify the assignable causes.

- Implement corrective actions to eliminate these causes.

Once the causes are addressed, any out-of-control points (samples) are removed from the chart.
Following this, revised center lines and control limits for the X̄-chart are recalculated based on the
remaining samples. These recalculated limits are called revised control limits.
¯
To calculate these revised limits, new values for X̄ new and R̄new are found using the following
formulas:

 
k d
¯ 1 X X
X̄ new = X̄i − X̄j  (23)
k − d i=1 j=1
 
k d
1 X X
R̄new = Ri − Rj  (24)
k − d i=1 j=1

¯
- X̄ new is the mean of the revised sample means after removing d samples from the original k
samples.
¯ , representing the average range of the
- R̄new is the revised range, calculated similarly to X̄new
remaining k − d samples.

¯
Revised Control Limits: Once X̄ new and R̄new are calculated, the center line and control
limits are updated using these new values:

28
 ¯
Central Line (CL) = X̄ (25)
new
¯
Lower Control Limit (LCL) = X̄new − A2 R̄new (26)
¯
Upper Control Limit (UCL) = X̄ + A R̄ (27)
new 2 new

When Using Sample Standard Deviation: If the process standard deviation is estimated
using the sample standard deviation, the center line and control limits are updated as follows:

¯
Central Line (CL) = X̄ (28)
new
¯
Lower Control Limit (LCL) = X̄new − A3 S̄new (29)
¯
Upper Control Limit (UCL) = X̄ + A S̄ (30)
new 3 new

where:  
k d
1 X X
S̄new = Si − Sj  (31)
k − d i=1 j=1

Pd
In this equation, j=1 Sj is the sum of standard deviations from the discarded samples.

Example:

A manufacturing plant monitors the diameter of bolts produced. For a specific machine, the
process mean (µ) and standard deviation (σ) for the diameter of the bolts are 12 mm and 0.4 mm,
respectively. The plant manager uses an X-bar chart with a subgroup size (n) of 6 and a constant
factor (A2 ) of 0.483.
Solution:
Given:

µ = 12 mm (process mean)
σ = 0.4 mm (process standard deviation)
n=6 (subgroup size)
A2 = 0.483 (constant factor)

29
 Using the formulas for the X-bar chart:

Center line (CL) = µ = 12 mm

σ
Upper control limit (UCL) = µ + A2 × √
n
σ
Lower control limit (LCL) = µ − A2 × √
n

Calculating the limits:

0.483 × 0.4
Upper control limit (UCL) ≈ 12 + √ ≈ 12 + 0.1932 ≈ 12.193 mm
6
0.483 × 0.4
Lower control limit (LCL) ≈ 12 − √ ≈ 12 − 0.1932 ≈ 11.807 mm
6

For this X-bar chart monitoring bolt diameters, the center line is 12 mm, the upper control
limit is approximately 12.193 mm, and the lower control limit is approximately 11.807 mm. These
control limits help determine whether the diameter of the bolts produced by the machine remains
within the expected quality range.

Example:

A packaging unit seals 50-gram packets of coffee. An X-bar control chart is used to monitor the
P
weight of these packets. After 30 samples, the sum of the X-values ( X) is 1550 grams, and the
P
sum of the R-values ( R) is 70 grams. Estimate the process mean and standard deviation, and
compute the control limits for the X-chart. Given that d2 = 2.059 and A2 = 0.483 for n = 5.

Solution:

Given:

- Number of samples, n = 5 (subgroup size for each sample).

P
- X = 1550 grams (sum of all X-values).
P
- R = 70 grams (sum of all R-values).

- d2 = 2.059 (for n = 5).

- A2 = 0.483 (for n = 5).

Calculations:
¯ ) is calculated as:
The mean of the sample means (X̄

30
 P
¯= X 1550
X̄ = = 51.67 grams
n 30
The mean of the ranges (R̄) is calculated as:
P
R 70
R̄ = = = 2.33 grams
n 30

Control Limits for the X-Chart:

The control limits are calculated using the given formulas:

¯ + A × R̄ = 51.67 + 0.483 × 2.33 = 51.67 + 1.12639 ≈ 52.80 grams

Upper Control Limit (UCL) = X̄ 2

¯ − A × R̄ = 51.67 − 0.483 × 2.33 = 51.67 − 1.12639 ≈ 50.54 grams

Lower Control Limit (LCL) = X̄ 2

Therefore, the control limits for the X-bar chart are:

- Upper Control Limit (UCL) = 52.80 grams.

- Lower Control Limit (LCL) = 50.54 grams.

- Centerline (CL) = 51.67 grams.

Example Data Recording Form for X-Chart:

Control Limits Based on Example Data:

¯)
To compute the control limits for the X-chart, we first calculate the mean of the sample means (X̄
and the mean of the ranges (R̄) based on the updated data.

¯ = 1 (5.60 + 5.70 + 5.70 + 5.82 + 5.65 + 5.80 + 5.75 + 5.75 + 5.85 + 5.70 + 5.75 + 5.82 + 5.80 + 5.65)
X̄
14
¯ ≈ 5.74
X̄

1
R̄ = (0.20 + 0.20 + 0.20 + 0.20 + 0.10 + 0.20 + 0.30 + 0.10 + 0.10 + 0.20 + 0.10 + 0.20 + 0.20 + 0.10)
14

R̄ = 0.18

31
 Table 4: Example Data Recording Form for X-Chart

Subgroup Date Time Measurements Mean R

1 10/01 09:00 5.6 5.5 5.7 5.6 5.60 0.20
2 10/01 10:30 5.7 5.8 5.7 5.6 5.70 0.20
3 10/01 12:00 5.6 5.7 5.7 5.8 5.70 0.20
4 10/01 13:30 5.9 5.8 5.7 5.9 5.82 0.20
5 10/01 15:00 5.6 5.7 5.6 5.7 5.65 0.10
6 10/01 16:30 5.8 5.9 5.8 5.7 5.80 0.20
7 10/01 18:00 5.6 5.7 5.8 5.9 5.75 0.30
8 10/01 19:30 5.7 5.8 5.8 5.7 5.75 0.10
9 10/01 21:00 5.8 5.9 5.8 5.9 5.85 0.10
10 10/02 10:30 5.7 5.8 5.7 5.6 5.70 0.20
11 10/02 12:00 5.8 5.7 5.8 5.7 5.75 0.10
12 10/02 13:30 5.7 5.8 5.9 5.9 5.82 0.20
13 10/02 15:00 5.8 5.9 5.8 5.7 5.80 0.20
14 10/02 16:30 5.6 5.7 5.6 5.7 5.65 0.10

Control Limits for the X-Chart

Now, we compute the control limits using A2 = 0.577 for n = 4:

¯ + A × R̄ = 5.74 + 0.577 × 0.18 = 5.74 + 0.10386 = 5.84

U CL = X̄ 2

¯ − A × R̄ = 5.74 − 0.577 × 0.18 = 5.74 − 0.10386 = 5.64

LCL = X̄ 2

Final Control Limits:

- Centerline (CL): 5.74

- Upper Control Limit (UCL): 5.84

- Lower Control Limit (LCL): 5.64

32
Graphical Representation:

Figure 7: Control Chart for X-Chart (Mean) with Centerline, UCL, and LCL.

Example
A novel manufacturing method for producing ball bearings has commenced. To oversee the
external diameter of these bearings, the quality controller gathers samples of five ball bearings at
10.00 AM, 12.00 PM, 2.00 PM, 4.00 PM, and 6.00 PM. The outside diameter of each chosen ball
bearing is measured in millimeters (mm). The outcomes of this assessment span a 4-day production
duration and are presented below. Determine and graph the central line and control limits for the
X-chart. Formulate conclusions regarding the process, considering special reasons for any points
outside the control limits. In the event that the process is not within control, compute the adjusted
central line and control limits.

33
 Table 5: Example Data Recording Form for X-Chart

Subgroup Date Time Measurements Mean R

X1 X2 X3 X4
1 Monday 10.00 AM 52 52 50 51 51.2 2
2 Monday 12.00 PM 50 53 52 53 51 3
3 Monday 2.00 PM 54 51 50 52 52 4
4 Monday 4.00 PM 56 55 53 55 54.4 3
5 Monday 6.00 PM 51 52 50 53 51.8 3
6 Tuesday 10.00 AM 50 52 51 50 51 2
7 Tuesday 12.00 PM 50 50 52 51 51.2 3
8 Tuesday 2.00 PM 52 51 53 50 51.2 3
9 Tuesday 4.00 PM 52 53 52 55 53 3
10 Tuesday 6.00 PM 51 51 50 51 51 2
11 Wednesday 10.00 AM 52 52 54 52 52.4 2
12 Wednesday 12.00 PM 49 48 50 50 49.6 3
13 Wednesday 2.00 PM 52 53 54 49 52 5
14 Wednesday 4.00 PM 52 51 54 51 52.4 3
15 Wednesday 6.00 PM 51 51 52 52 51.4 1
16 Thursday 10.00 AM 50 50 51 52 51 2
17 Thursday 12.00 PM 50 51 53 51 51.6 3
18 Thursday 2.00 PM 52 50 49 53 50.8 4
19 Thursday 4.00 PM 52 51 54 51 51.8 3
20 Thursday 6.00 PM 51 51 50 52 51.2 2
Total 1032.4 56

Solution
In the absence of knowledge regarding the process average (µ) and process variability (σ), we
employ the following equations to establish the center line and control limits for the X-chart. A2
= 0.577 for n = 5, based on the information provided previously.

¯= 1 1
X
X̄ X̄ = × 1032.4 = 51.62
k 20
1X 1
R̄ = R= × 56 = 2.80
k 20
¯ = 51.62
CL = µ̂ = X̄
3 ¯ − √3 R̄ = 51.62 − 0.577 × 2.80 = 50.004
LCL = µ̂ − √ σ̂ = X̄
n n d2
3 ¯ + √3 R̄ = 51.62 + 0.577 × 2.80 = 53.236
UCL = µ̂ + √ σ̂ = X̄
n n d2
We proceed to create the X-chart, depicting the sample number on the X-axis and the average
diameter of the ball bearing (X) on the Y-axis, illustrated in Figure 9.

34
 Figure 8: The X̄-chart for the average diameter of the ball bearings.

From Fig. 9, we observe that the points corresponding to samples 4 and 12 lie outside the control
limits. Therefore, the process is out-of-control and some assignable causes are present in the process.
To bring the process under statistical control, it is necessary to investigate the assignable causes
and take corrective action to eliminate them.
After eliminating the assignable causes from the process, we delete the out-of-control points (4
and 12 samples) and calculate the revised centre line and control limits for the X̄-chart using the
remaining samples. For revised limits of the X̄-chart, we first calculate the new X̄ and new R̄:

k = 20

d=2
d
X
X̄j = 54.4 + 49.6 = 104
j=1

d
X
Rj = 3 + 3 = 6
j=1
 
k d
¯ 1 X X 1032.4 − 104
X̄ new = X̄i − X̄j  = = 51.578
k − d i=1 j=1
20 − 2
 
k d
1 X X 56 − 6
R̄new = Ri − Rj  = = 2.778
k − d i=1 j=1
20 − 2

35
 Now, we calculate the revised centre line and control limits of the X̄-chart as follows:

¯
CL = X̄ new = 51.578

¯
LCL = X̄ new − A2 R̄new = 51.578 − 0.577 × 2.778 = 49.975

¯
U CL = X̄ new + A2 R̄new = 51.578 + 0.577 × 2.778 = 53.181

Figure 9: The revised X̄-chart for the average diameter of the ball bearings.

2.7 Control of Process Variability: Range Chart (R-Chart)

The fundamental logic and structure of an R-chart closely resemble that of the X-chart. However,
the key distinction lies in what is plotted: instead of tracking sample means, the R-chart focuses
on sample ranges, monitoring the variability in the process.
The range (R) in a single sample is calculated as the difference between the maximum and
minimum values within that sample:

R = Xmax − Xmin (32)

36
Understanding Control Limits for the R-Chart

To establish control limits for the R-chart, a fundamental understanding of the sampling distribution
of the range is required. While a detailed derivation of this distribution is beyond the scope of this
course, the key results can be presented directly.
For a random sample X1 , X2 , . . . , Xn drawn from a normal population with mean µ and variance
2
σ , the mean and variance of the sampling distribution of the range (R) are given as follows:

E(R) = d2 σ (33)

Var(R) = d23 σ 2 (34)

Where:

- E(R) is the expected (average) value of the range,

- Var(R) is the variance of the range,

- d2 and d3 are constants that depend on the sample size n,

- σ is the process standard deviation.

Interpretation of E(R) and V ar(R)

The formulas for E(R) (the expected value of the range) and Var(R) (the variance of the range)
describe the average behavior of the range across multiple samples drawn from the same population.
- E(R) represents the average range you would expect to observe in many samples drawn from
the population.
- Var(R) represents how much the observed range values are expected to fluctuate from sample to
sample.
While in any single sample, R is simply the observed difference between the largest and smallest
values, E(R) and Var(R) provide the theoretical framework that allows us to set up control limits
for monitoring the variability in a process over time.

Control Limits for the R-Chart

The control limits for the R-chart are established using the expected value of the range and its
variability, based on the process standard deviation and constants derived from statistical tables.
These limits enable us to monitor whether the process variability is within acceptable bounds or
whether special causes of variation are affecting the process.

37
 Thus, by constructing an R-chart, we can effectively track the spread of data within a sample
and identify when a process becomes unstable due to excessive variability. This makes the R-chart
a powerful tool in maintaining process control and ensuring consistent product quality.
Here, d2 and d3 denote constants dependent on the sample size, and data providing these
constants for various sample sizes can be found in Table 6.

Table 6: Values for Control Chart Constants (d2 , d3 , D1, D2, D3 , D4 , B5 , B6 )

Sample
d2 d3 D1 D2 D3 D4 B5 B6
Size (n)
2 1.128 0.853 0.000 3.686 0 3.267 0 2.606
3 1.693 0.888 0.000 4.358 0 2.574 0 2.276
4 2.059 0.880 0.000 4.698 0 2.282 0 2.088
5 2.326 0.864 0.000 4.918 0 2.114 0 1.964
6 2.534 0.848 0.000 5.078 0 2.004 0.029 1.874
7 2.704 0.833 0.000 5.204 0.076 1.924 0.113 1.806
8 2.847 0.820 0.204 5.306 0.136 1.864 0.179 1.751
9 2.970 0.808 0.388 5.393 0.184 1.816 0.232 1.707
10 3.078 0.797 0.547 5.469 0.223 1.777 0.276 1.669
11 3.173 0.787 0.687 5.535 0.256 1.744 0.313 1.637
12 3.258 0.778 0.811 5.594 0.283 1.717 0.346 1.610
13 3.336 0.770 0.922 5.647 0.307 1.693 0.374 1.585
14 3.407 0.763 1.025 5.696 0.328 1.672 0.399 1.563
15 3.472 0.756 1.118 5.741 0.347 1.653 0.421 1.544
16 3.532 0.750 1.203 5.782 0.363 1.637 0.440 1.526
17 3.588 0.744 1.282 5.820 0.378 1.622 0.458 1.511
18 3.640 0.739 1.356 5.856 0.391 1.608 0.475 1.496
19 3.689 0.734 1.424 5.891 0.403 1.597 0.490 1.483
20 3.735 0.729 1.487 5.921 0.415 1.585 0.504 1.470

Note The statistical rationale behind this omission is that a single data point (sample size
of 1) does not allow the calculation of a range or other measures, as there is no variation to
compute within a single observation. There is no range without multiple values. Therefore, for
control chart calculations, it is not practical or statistically meaningful to consider a sample size of
one. Additionally, in quality control practices, control charts are predominantly used to monitor
processes based on subgroups or samples. Using individual data points (sample size of 1) would not
capture inherent variability in the process, making control limits irrelevant for such small sample
sizes. As a result, these control chart constant tables typically start from sample sizes greater
than one, where there is enough data to compute measures of variability or distribution that are
meaningful for control chart construction and process monitoring.
p
Recognizing that SE(X) = V ar(X), we derive the standard error of the range:

38
 p q
SE(R) = V ar(R) = d23 σ 2 = d3 σ (21)

While the sampling distribution of the range does not adhere to a normal distribution, it is
common practice to utilize ±3σ limits. These limits ensure that the probability of an observation
falling outside these bounds is exceedingly low. Consequently, the control limits (CL, LCL, and
UCL) for the R-chart are defined as follows:

Center Line (CL): CL = E(R) = d2 σ (35)

Lower Control Limit (LCL): LCL = E(R) − 3SE(R) =
d2 σ − 3d3 σ = (d2 − 3d3 )σ = D1σ (36)
Upper Control Limit (UCL): U CL = E(R) + 3SE(R) =
d2 σ + 3d3 σ = (d2 + 3d3 )σ = D2σ (37)

Here, constants D1 = d2 − 3d3 and D2 = d2 + 3d3 are size-dependent, and their values can be
referenced in Table 6.

2.8 Estimating σ for the R-Chart

In practical scenarios, the true value of σ (process standard deviation) is often unknown. Therefore,
it is estimated from samples, which are taken when the process is believed to be under statistical
control. For the R-chart, particularly when dealing with small sample sizes (< 10), σ is estimated
using the sample range (R). If X1 , X2 , . . . , Xn represent a sample of size n, σ is estimated from the
following equation:

R̄
σ̂ = (38)
d2
Where:

- σ̂ is the estimated standard deviation,

- R̄ is the average sample range, and

- d2 is a constant that depends on the sample size.

Control Limits for the R-Chart

R̄
The center line and control limits for the R-chart, when σ is estimated by d2 , are given as:

39
 - Center Line (CL):
CL = R̄ (39)

The center line is simply the average of the sample ranges (R̄).

- Lower Control Limit (LCL):

3d3
LCL = R̄ − 3SE(R) = R̄ − R̄
d2
 
3d3
LCL = 1− R̄ = D3 R̄ (40)
d2
 
3d3
Where D3 = 1 − d2 is a constant that depends on the sample size.

- Upper Control Limit (UCL):

3d3
UCL = R̄ + 3SE(R) = R̄ + R̄
d2
 
3d3
UCL = 1 + R̄ = D4 R̄ (41)
d2
 
3d3
Where D4 = 1 + d2 is another constant that depends on the sample size.

Thus, the control limits for the R-chart are computed using constants D3 and D4 , which vary
based on the sample size n. These constants have been tabulated for various sample sizes to simplify
the calculation of control limits.

2.9 Control of Process Variability: Range Chart (R-Chart)

Once the center line and control limits are established, the R-chart is created by plotting the sample
number on the X-axis (horizontal scale) and the sample range (R) on the Y-axis (vertical scale).
The center line is represented by a solid line, while the upper control limit (UCL) and lower control
limit (LCL) are depicted by dotted lines. The sample range values for each sample are plotted
against the sample number, with consecutive points connected by line segments.
Interpreting the results involves assessing the positioning of sample points relative to the control
limits. If all sample points fall within or on the upper and lower control limits, the R-chart signals
that the process is under statistical control. Conversely, if points lie outside these limits, the process
is considered out of control. To rectify this situation, a thorough investigation of assignable causes
is necessary, followed by corrective action to eliminate them. Out-of-control points are removed,

40
and a revised center line and control limits are calculated, mirroring the process used for the control
chart of process mean. The revised limits for the R-chart are defined as follows:

CL = R̄new (42)
LCL = D3 R̄new (43)
U CL = D4 R̄new (44)

Key Considerations When Using X̄ and R-Charts Together

When employing X̄ and R-charts, several key considerations should be kept in mind:

- To ascertain if a process is under statistical control:

- Monitoring both the process mean and process variability is crucial.

- The R-chart is usually analyzed before the X̄-chart. The R-chart provides insights into
the variability within the individual samples or subgroups. If there are indications of
excessive variation in the R-chart (for example, if points fall outside control limits), it
could suggest issues with the measurement system or instability in the process. Address-
ing and stabilizing this variation can be a priority before further analyzing the subgroup
averages.
- The R-chart focuses on the variability in the process, providing insights into the consis-
tency of sample ranges.
- However, it is important to note that the order of assessment does not necessarily have
to be fixed. The R-chart and X̄-bar chart complement each other, and they often work
in tandem to provide a comprehensive understanding of process variation. The choice
of which chart to prioritize first may depend on the specific context of the process, the
nature of issues observed, and the goals of process improvement. In practice, both charts
are often reviewed and interpreted simultaneously to gain a complete understanding of
process behavior.

- If the R-chart indicates out-of-control process variability:

- Corrective actions are initiated to eliminate assignable causes that contribute to the
observed variability.
- Achieving statistical control with the R-chart is a prerequisite before analyzing the X̄-
chart.
- Bringing the R-chart under control automatically addresses many assignable causes that
might affect the X̄-chart, ensuring a more accurate assessment of the process mean.

41
 - If the R-chart indicates in-control process variability but X̄-chart signals an out-
of-control process mean:

- The R-chart continues to be used as it provides valuable information on process variabil-

ity.
- The X̄-chart is revised by eliminating points that indicate an out-of-control process
mean.
- The center line and control limits of the X̄-chart are adjusted based on the remaining
data.
- Despite this adjustment, the distance between control limits remains the same as in the
out-of-control scenario. This is because the sample range (used for the R-chart) is not
recalculated; only the problematic points are removed.

- When both X̄ and R-charts indicate statistical control:

- The combined use of X̄ and R-charts provides a comprehensive assessment of the process.
¯ (grand mean of X̄) and R̄ can be considered
- If both charts indicate statistical control, X̄
representative of the process.
- Their values are taken as standard, suggesting that the process is stable, and variations
are within acceptable limits.

Practical Example: Combined Use of X̄ and R-Charts

Let’s consider a practical example where a factory produces metal components with a specified
diameter. A quality control team monitors the diameter using both X̄ and R-charts to ensure
the process is under control. The R-chart focuses on the variability between samples, whereas the
X̄-chart monitors the average diameter over time.
After plotting both charts, the R-chart indicates all sample ranges fall within control limits,
but the X̄-chart shows several points outside the upper control limit, suggesting an issue with the
process mean. Upon investigation, the team identifies a minor calibration issue with the measuring
equipment. After correcting the issue, they remove the out-of-control points from the X̄-chart and
recalculate the center line and control limits.
This example illustrates how both charts work together to provide a complete picture of the
process, highlighting not only issues with variability but also with the process mean.
Example:
A statistical quality controller uses the X̄ and R-charts together for monitoring the quality
characteristic of a product. Samples of size n = 5 are taken from the manufacturing process at
regular intervals. A normally distributed quality characteristic is measured, and the X̄ and R are

42
 ¯ = 6.40 and R̄ = 0.0877.
calculated for each sample. After 20 samples have been analyzed, we have X̄
Compute the center line and control limits for the X̄ and R-charts.
Solution:
The average (µ) and variation (σ) of the process are unknown in this situation. Hence, we employ
the following equations to determine the central value and control boundaries for the X-chart and
R-chart, respectively.
Given data:

¯ = 6.40
X̄
R̄ = 0.0877
n=5
k = 20

Constants:

A2 = 0.577
D3 = 0 (since D3 = 0 for small samples)
D4 = 2.114

X̄-chart Center Line and Control Limits:

¯ = 6.40
CL = X̄
¯ + A × R̄ = 6.40 + 0.577 × 0.0877 = 6.451
U CL = X̄ 2
¯
LCL = X̄ − A × R̄ = 6.40 − 0.577 × 0.0877 = 6.349
2

R-chart Center Line and Control Limits:

CL = R̄ = 0.0877
U CL = D4 × R̄ = 2.114 × 0.0877 = 0.185
LCL = D3 × R̄ = 0 × 0.0877 = 0

These control limits help to assess whether the process is under statistical control. The R-chart
shows whether the variability in the process is stable, while the X̄-chart monitors whether the
process mean is consistent.
Example
A dairy company employs automated machines to package 500 ml of milk. A quality control

43
inspector examines four packets in each sample at regular intervals, measuring the weight of each
filled packet. The dataset for 20 samples is presented in the table below:

Sample Weight of Filled Milk Packet (in ml)

X̄ R
X1 X2 X3 X4
1 500 520 500 500 506.67 20
2 500 490 520 530 503.33 40
3 490 550 570 540 536.67 80
4 510 520 500 520 510.00 20
5 510 480 490 490 493.33 30
6 520 500 520 500 513.33 20
7 520 510 530 510 520.00 20
8 530 490 520 500 513.33 40
9 510 490 500 510 500.00 20
10 520 520 490 520 510.00 30
11 520 500 510 500 510.00 20
12 480 500 520 510 500.00 40
13 530 510 520 510 520.00 20
14 500 510 510 500 506.67 10
15 490 520 500 510 503.33 30
16 520 500 530 500 516.67 30
17 520 560 490 510 523.33 70
18 500 490 500 510 496.67 20
19 520 500 530 500 516.67 30
20 500 490 500 490 496.67 10
Total 10196.67 600

Utilizing both the X̄ and R-charts simultaneously, draw inferences about the process by con-
sidering assignable causes for any points that fall out of control. In the event that the process is
deemed out-of-control, compute the revised center line and control limits to restore the process to
a state of statistical control.
Solution
Since the X̄ and R-charts are intended for combined use, we initiate our analysis with the
R-chart.
In this context, the variability of the process (σ) remains unknown. Hence, we employ equations
(39 to 41) to compute the center and control limits for the R-chart. We acquire the values A2 =
0.729, D3 = 0, and D4 = 2.282 for n = 4.
Our initial step involves calculating the value of R for the center line and control limits based
on the provided data, utilizing equation as outlined below:

1X 1
R̄ = R= × 600 = 30
k 20
By substituting the values of R̄, D3 , and D4 into equations (39 to 41), we derive:

44
 CL = R̄ = 30

UCL = D4 R̄ = 2.282 × 30 = 68.46

LCL = D3 R̄ = 0 × 30 = 0

We subsequently proceed to construct the R-chart, employing the sample number on the X-axis
and the sample range (R) of the milk packets on the Y-axis, as illustrated in Fig. 10.

Figure 10: The R-chart for milk packets.

Interpretation of the Result From Fig. 10, it is evident that the points corresponding to
samples 3 and 17 lie outside the control limits, indicating that the process variability is out-of-
control with the presence of assignable causes. To calculate the revised control limits, the out-of-
control points are removed, and the new R̄new for the remaining samples is computed. For this
Pd
example, where k = 20, d = 2, and j=1 Rj = 80 + 70 = 150:
Pk Pd
i=1 Ri − j=1 Rj 600 − 150 450
R̄new = = = = 25
k−d 20 − 2 18
The revised center line and control limits of the R-chart are then calculated using equations:

45
 CL = R̄new = 25 (45)

UCL = D4 R̄new = 2.282 × 25 = 57.05 (46)

LCL = D3 R̄new = 0 × 25 = 0 (47)

Fig. 11 illustrates the revised R-chart.

Figure 11: The revised R-chart.

Analysis of the Results Upon reviewing the updated R-chart, illustrated in Fig. 11, it is evident
that all data points fall within the control limits. This indicates that the process variability is now
within statistical control. It is important to highlight that if any points had remained outside the
revised control limits, further adjustments would be necessary until achieving statistical control.
Following the resolution of process variability issues, the focus shifts to the examination of
the process mean. Control limits for the X̄-chart are computed for the remaining samples. The
¯ value with k = 20, d = 2, and Pd X̄ = 536.67 + 523.33 = 1060, is:
recalculated X̄ j=1 j

Pk Pd
¯ j=1 X̄i − j=1 X̄j 10196.67 − 1060 9136.67
X̄ new = = = = 507.593
k−d 20 − 2 18
Subsequently, the updated center line and control limits for the X̄-chart are calculated using
equations (16a to 16c):

46
 Figure 12: The X̄-chart for milk packets.

¯
CL = X̄ new = 507.593 (48)

¯
UCL = X̄ new + A2 R̄new = 507.593 + 0.729 × 25 = 525.818 (49)

¯
LCL = X̄ new − A2 R̄new = 507.593 − 0.729 × 25 = 489.368 (50)

The X̄-chart can be effectively visualized by plotting the sample numbers on the X-axis and the
average weight of milk packets on the Y-axis, as depicted in Fig. 12.

Interpretation of the Results Upon careful examination of the updated X̄-chart, as presented
in Fig. 12, it is notable that none of the data points fall outside the control limits. This observation
signifies that the process mean is within statistical control.

2.10 Understanding the Standard Deviation Chart (S-Chart)

In previous lessons, we explored control charts for monitoring process mean (X-chart) and process
variability (R-chart). While the R-chart is straightforward, it falls short for larger sample sizes
(typically exceeding 10). The limitation arises because the range (R-chart) depends solely on the
maximum and minimum values, whereas the standard deviation considers all observations. As the
number of observations increases, the R-chart’s fluctuation becomes more pronounced compared

47
to the standard deviation. For sample sizes of 10 or more, a standard deviation chart (S-chart)
becomes more suitable than the range chart (R-chart).
To establish control limits for the S-chart, understanding the sampling distribution of the stan-
dard deviation (S) is essential. As the derivation of this distribution is beyond our course’s scope,
we present the result without proof. If X1 , X2 , . . . , Xn constitutes a random sample of size n from
a normal distribution with mean µ and variance σ 2 , then:

E(S) = c4 σ

V ar(S) = (1 − c24 )σ 2

Here, c4 is a constant dependent on the sample size, and its values are tabulated in Table 6.
Additionally, it is known that:
p
SE(X) = V ar(X)
p q
SE(S) = V ar(S) = σ 1 − c24

While the sampling distribution of the standard deviation is not normal, a common practice is
to use ±3σ limits. This choice ensures that the probability of an observation falling outside these
limits is minimal. Thus, the CL, LCL, and UCL for the S-chart are computed as follows:

Centre line (CL): E(S) = c4 σ (51)

q
Lower control limit (LCL): E(S) − 3SE(S) = c4 σ − 3σ 1 − c24
q
LCL = (c4 − 3 1 − c24 )σ = B5 σ (52)
q
Upper control limit (UCL): E(S) + 3SE(S) = c4 σ + 3σ 1 − c24
q
U CL = (c4 + 3 1 − c24 )σ = B6 σ (53)
p p
Here, B5 = (c4 − 3 1 − c24 ) and B6 = (c4 + 3 1 − c24 ) are constants contingent on the sample
size, and their values can be found in Table 6.
In practical scenarios, the true value of σ is often unknown. Consequently, it needs to be
estimated using samples taken when the process is believed to be under statistical control. As the
S-chart is designed for large sample sizes (≥ 10), σ is approximated using the sample standard
deviation (S).
The control limits for the S-chart, considering the estimation of σ through the average sample
standard deviation, are expressed as follows:

- Centre line (CL): The center line remains the same as E(S), calculated using c4 and the

48
 estimated σ.

- Lower control limit (LCL): The lower control limit is given by E(S) − 3SE(S), where
p
SE(S) involves the estimated σ. This is equivalent to (c4 − 3 1 − c24 )S.

- Upper control limit (UCL): The upper control limit is given by E(S) + 3SE(S), incorpo-
p
rating the estimated σ. This is expressed as (c4 + 3 1 − c24 )S.

In these equations, c4 still represents the constant dependent on the sample size, and the esti-
mated σ is replaced by the sample standard deviation (S). These adjustments cater to the practical
scenario where the true value of σ is not known, and an estimate is used instead.

S̄
CL = E(S) = c4 σ̂ = c4 = S̄ (54)
c4
q
LCL = E(S) − 3SE(S) = c4 σ̂ − 3σ̂ 1 − c24

 
S̄ S̄ 3
q q
or LCL = c4 − 3 2
1 − c4 = 1 − 2
1 − c4 S̄ = B3 S̄ (55)
c4 c4 c4
q
UCL = E(S) + 3SE(S) = c4 σ̂ + 3σ̂ 1 − c24

 
S̄ S̄ 3
q q
or UCL = c4 +3 1 − c24 = 1+ 1 − c24 S̄ = B4 S̄ (56)
c4 c4 c4
The creation of the S-chart mirrors that of the R-chart, employing a similar framework. The
X-axis represents the sample number, while the Y-axis illustrates the sample standard deviation. A
solid line signifies the center line, and dotted lines delineate the upper and lower control limits. Each
sample’s standard deviation value is graphically depicted against the sample number, connected by
line segments.
Interpretation of the result: A well-behaved S-chart reveals that all sample points lie within
the upper and lower control limits, signifying that the process is statistically under control. Devi-
ations beyond these limits indicate an out-of-control process, demanding a thorough investigation
into assignable causes. To rectify this, out-of-control points are excluded, and revised center lines
and control limits—referred to as revised control limits—are computed.
For the revised limits of the S-chart, compute S̄new using the formula:
Pk Pd
i=1 Si − j=1 Sj
S̄new =
k−d

49
 Pd
Here, d signifies the number of discarded samples, and j=1 Sj is the sum of standard deviations
of the excluded samples.
Reconstruct the center line and control limits of the S-chart by substituting S with S̄new in the
subsequent equations:

Center Line (CL) = S̄new (57)

Lower Control Limit (LCL) = B3 S̄new (58)
Upper Control Limit (UCL) = B4 S̄new (59)

This meticulous process ensures a nuanced comprehension of the process’s variability and fa-
cilitates the maintenance of statistical control. The subsequent example will further elucidate the
intricate steps involved in constructing and interpreting the S-chart.
Example The table provides the mean and standard deviation of 15 samples, each consisting
of 12 data points, from a production process. Evaluate the control limits for both the X̄-chart
and S-chart. Provide insights into whether the process variability and mean are within statistical
control. Utilize quality control factors A3 = 0.886, B3 = 0.354, B4 = 1.646 for n = 12.
Given Data:
Sample No Mean Standard Deviation
1 9.5 1.2
2 7.5 1.5
3 10.0 2.6
4 8.25 3.2
5 7.25 1.9
6 9.0 2.6
7 10.5 5.4
8 9.25 2.4
9 8.75 2.1
10 9.75 3.5
11 7.25 2.6
12 10.25 1.1
13 9.25 2.2
14 10.0 1.6
15 10.5 3.2
Solution: In this case, our objective is to manage both process variability and process mean.
As the sample size exceeds 10, we employ S and X-charts. Given the sample size n > 10, we apply
equations (20 to 22) and (54 to 56) to compute the central tendencies and control limits for the
X-chart and S-chart, respectively.

50
 Control Limits for X-Chart:
k
¯= 1 1
X
X̄ X̄i = (9.5 + 7.5 + . . . + 10.5) = 9.133
k i=1 15
k
1X 1
S̄ = Si = (1.2 + 1.5 + . . . + 3.2) = 2.473
k i=1 15

To verify the statistical control of the process, we initially construct and examine the S-chart.
By substituting the values of S̄, B3 , and B4 into equations (29a to 29c), we determine the central
line and control limits of the S-chart as follows:
Control Limits for S-Chart:

CL = S̄ = 2.473
U CL4 = B4 S̄ = 1.646 × 2.473 = 4.071
LCL3 = B3 S̄ = 0.354 × 2.473 = 0.875

We create the S-chart by plotting the sample number on the X-axis and representing the sample
standard deviation (S) on the Y-axis, as illustrated in Figure 13.

Figure 13: The S-chart for Example.

Interpretation of the result The S-chart illustrated in Figure 13 reveals that the data point
corresponding to sample 7 exceeds the upper control limits. This indicates that the process vari-
ability is out-of-control, suggesting the presence of assignable causes in the process. To restore the
process to statistical control, it is essential to investigate and address these assignable causes. By

51
discarding sample 7 we have:
Pk Pd
i=1 Si − j=1 Sj 37.095 − 5.4
S̄new = = = 2.263
k−d 15 − 1
Then we can generate new control limits as:

CL = S̄new = 2.263
U CL4 = B4 S̄new = 1.646 × 2.263 = 3.726
LCL3 = B3 S̄new = 0.354 × 2.263 = 0.801

As you can see, the remaining samples remain between LCL and UCL by pruning sample 7 from
the process.
Next, we calculate the center line and control limits for the X̄-chart. Before discarding sample
7 the control limits were as follows:

¯ = 9.133
CL = X̄
¯ + A S̄ = 9.133 + 0.886 × 2.473 = 11.324
UCL = X̄ 3
¯
LCL = X̄ − A S̄ = 9.133 − 0.886 × 2.473 = 6.942.
3

We construct the X-chart by plotting the sample number on the X-axis and the sample average
(X) on the Y-axis, as depicted in Figure 14.

Figure 14: The X̄-chart for Example.

Interpretation of the result The X̄-chart depicted in Figure 14 reveals that all sample points
fall within the control limits. Consequently, it indicates that the process mean was under statistical

52
control before discarding sample 7.
Therefore, by utilizing both the S and X̄-charts, we can deduce that the process variability is
not under statistical control, while the process mean is indeed under statistical control.
Homework Attempt to redefine the control limits for the X̄-chart by excluding sample size 7
from the dataset. Evaluate whether the process remains within control limits or not.

2.11 PROCESS CAPABILITY ANALYSIS

Variation in manufactured products is an inherent aspect of industrial processes, stemming from
diverse sources of natural variation. Even when a production process is well-designed or meticulously
maintained, it is practically impossible for two items or units to be identical. The differences between
items can range from significant to subtle, depending on the sources of variation.
In certain cases, the process might be under statistical control, yet customers remain dissatisfied
due to non-conformity with product specifications. This situation necessitates a process capabil-
ity analysis to assess whether a product aligns with its specified requirements. For example, in a
manufacturing setting, a company produces bolts designed to meet a specific length requirement of
20 mm. The production process has been meticulously controlled, and statistical analysis indicates
that the process is under control according to control charts for bolt length. However, despite the
process being statistically under control, the company receives complaints from customers about
bolts not meeting the specified length requirement. This discrepancy prompts a deeper investi-
gation as to why the produced bolts, although within statistical control, fail to satisfy customer
expectations regarding product specifications.
To initiate a process capability study, it is crucial to differentiate between specification limits,
control limits, and natural tolerance limits.

2.11.1 Specification Limits

Specification limits are externally determined boundaries set by consumers, management, or man-
ufacturing engineers during the design and development stages of a product. These limits define
the acceptable range for a variable product characteristic. The upper specification limit (USL) is
the maximum allowable value, while the lower specification limit (LSL) is the minimum. A product
falling outside these limits is considered defective and may need to be scrapped or reworked.
Example: In the automotive industry, consider the manufacturing of engine pistons. During the
design phase, engineers establish specification limits for the piston diameter, crucial for ensuring
optimal engine performance. The upper specification limit (USL) is set at 75.2 millimeters to
prevent the piston from causing excessive friction within the cylinder, while the lower specification
limit (LSL) is set at 74.8 millimeters to ensure adequate contact and sealing. Any piston diameter
falling below 74.8 millimeters or exceeding 75.2 millimeters would be considered defective as it

53
could compromise engine efficiency or potentially cause damage. Pistons outside these limits might
lead to increased friction, decreased fuel efficiency, or even engine malfunction. In the production
process, strict quality control measures are implemented to ensure that the manufactured pistons
fall within these defined specification limits. Quality inspectors regularly measure piston diameters,
and any pistons found outside the specified range are deemed defective. These defective pistons
might undergo rework, or if unable to meet specifications after rework, they are scrapped to maintain
the required quality standards and ensure optimal engine performance for end-users.
Example: Consider the weight of a cricket ball specified as 160.0 ± 3.0 grams. The USL is
163.0 grams, and the LSL is 157.0 grams. Any ball produced outside this weight range may be
deemed defective.

2.11.2 Control Limits

We discussed that control limits for a process are established through the collection and analysis
of samples. These limits act as thresholds, and if a data point from a sample falls beyond these
limits, it signals that the process is potentially out of control. The purpose of these control limits
is to provide a visual and statistical indication of the stability of the manufacturing process.
It is important to note that if an individual product measurement surpasses these control limits,
it does not automatically imply that the entire process is defective. This is a key distinction. The
reason lies in the nature of control charts, which typically depict aggregate statistics such as the
sample mean or range, rather than individual measurements.
By focusing on summary statistics, control charts provide a more comprehensive view of the
overall process behavior. This approach safeguards against overreacting to occasional, isolated
deviations in individual product measurements. A single product falling outside the control limits
might not be a cause for immediate concern, as the chart considers the collective behavior of the
process over time.
In essence, control charts offer a balanced perspective, allowing for the identification of sustained
patterns of variation rather than reacting to isolated events. This methodology contributes to a
more stable and controlled manufacturing process, minimizing unnecessary interventions triggered
by isolated occurrences.

2.11.3 Natural Tolerance Limits

Natural tolerance limits (NTLs) serve as benchmarks determined by individual measurements of

a product’s characteristic. These limits capture the intrinsic variability of products within the
manufacturing process and are positioned at the 3σ range from the mean (µ ± 3σ). Specifically,
the upper natural tolerance limit (UNTL) corresponds to µ + 3σ, while the lower natural tolerance
limit (LNTL) is expressed as µ − 3σ. In instances where µ and σ are unknown, their values can be

54
approximated using the grand sample mean and range or standard deviation.
It is crucial to recognize that although specification limits are externally defined by customers,
control limits and natural tolerance limits are intrinsically linked to process variability. There exists
a meaningful relationship among them, which can be clarified through a practical example.
Example: Consider a manufacturing process where the specification limits for a product are set
at USL = 105 and LSL = 95. The process is under control, and the control limits are determined
to be UCL = 103 and LCL = 97. Additionally, the natural tolerance limits, calculated as µ ± 3σ,
are found to be UNTL = 102 and LNTL = 98. In this scenario, we observe that the control
limits are slightly wider than the natural tolerance limits, indicating the process is well within
statistical control. However, there is a narrow margin between the natural tolerance limits and
the specification limits, suggesting that even slight process variations could lead to non-conforming
products. To mitigate this risk, the process could be further optimized or adjusted to align the
process mean closer to the center of the specification range.
In this example, the relationships between control limits, natural tolerance limits, and specifica-
tion limits highlight the importance of continuous monitoring and improvement. Process capability
analysis helps determine whether the process can consistently meet customer specifications and
maintain product quality.
Example: Control of Cricket Ball Weight
A cricket ball manufacturing company wants to control the weight of the ball. Twenty-five
samples, each of size 4 were collected. The sum of averages and sum of sample ranges were found to
be 4010 grams and 72 grams, respectively. If the customer specified the weight of the ball as 160±3.0
grams, find the estimates of process mean and standard deviation. Also find the specification limits,
control limits, and natural tolerance limits. It is given that, d2 = 2.088 and A2 = 0.729 for n = 4.
Detailed Solution

¯ ) and Standard Deviation (S̄):

1. Estimates of Process Mean (X̄
¯ ) is obtained by dividing the sum of averages by the number of
The average of averages (X̄
samples:
¯ = 4010 = 160.4
X̄
25

Similarly, the average of sample ranges (R̄) is calculated as:

72
R̄ = = 2.88
25

The estimates for the process mean (µ) and standard deviation (σ) are then given by:

¯ = 160.4, R̄ 2.88
µ̂ = X̄ σ̂ = = ≈ 1.38
d2 2.088

55
 2. Specification Limits:
The customer-specified weight range is 160 ± 3.0 grams. Therefore, the upper specification
limit (USL) is 160 + 3.0 = 163.0 grams, and the lower specification limit (LSL) is 160 − 3.0 =
157.0 grams.

3. Control Limits:
Using the given A2 value and the estimated R̄, control limits for the process are determined:

¯ = 160.4,
CL = X̄

¯ + A R̄ = 160.4 + (0.729 × 2.88) ≈ 162.4

UCL = X̄ 2

¯ − A R̄ = 160.4 − (0.729 × 2.88) ≈ 158.4

LCL = X̄ 2

4. Natural Tolerance Limits (NTLs):

The NTLs, representing 3σ from the mean, are calculated as (if µ and σ are unknown,
estimates are used):
   
¯ +3 R̄ 2.88
UNTL = X̄ ≈ 160.4 + 3 ≈ 164.16
d2 2.088
   
¯ −3 R̄ 2.88
LNTL = X̄ ≈ 160.4 − 3 ≈ 156.64
d2 2.088

Process Capability Analysis

Process capability is a metric that assesses the ability of a manufacturing process to produce items
within specified limits. While statistical control provides insights into the stability of a process,
process capability analysis determines whether the process can consistently meet specific product
specifications. It measures the spread of a quality characteristic when the process is stable.
If the standard deviation of a stable process is denoted as σ, process capability is defined as 6σ,
indicating that products can be manufactured within ±3σ limits. In practice, the true standard
deviation σ is often unknown, and we estimate it using either the range R̄ or the standard deviation
S̄, calculated with appropriate constants like d2 or c4 .
To evaluate whether the process is capable of meeting specific limits, we compare process capa-
bility with tolerance, where tolerance is the difference between specification limits:

Tolerance = USL − LSL

Three cases can arise from this comparison:

56
 - Case I: 6σ < USL − LSL
In this scenario, the process is capable of producing products within the desired specifica-
tions, irrespective of whether the process is under statistical control or out-of-control. There
is sufficient room within the tolerance range to accommodate the process variability, ensur-
ing products consistently meet specifications. When 6σ (representing process variability) is
smaller than the tolerance (representing the acceptable range), it implies that the process can
consistently produce outputs that fall comfortably within the specified limits.

- Case II: 6σ = USL − LSL

Here, the process can meet specifications only if there is no shift in the process mean or a
change in process variability. Even a slight shift in the process could cause products to exceed
the tolerance boundaries. It becomes crucial to maintain the process stability and prevent
changes in mean or variation to meet specifications consistently.

- Case III: 6σ > USL − LSL

In this situation, the products do not meet the desired specifications, even if the process is
under statistical control. The inherent variability in the process exceeds the tolerance range,
indicating that the process, as it stands, cannot consistently produce products within the
specified limits. Therefore, reducing process variability is necessary to meet the specifications.

The relationship between process capability and specification limits can be illustrated with a
common example from daily life.
Scenario 1: Car Slightly Smaller Than Garage

- Representation: The car is slightly smaller than the garage.

- Analogy: Parking a slightly smaller car in the garage aligns with the midpoint of specifica-
tions. Precision in centering is essential to align with the specified boundaries.

- Interpretation: In a process with minimal variability (similar to the car fitting almost
perfectly within the garage), manufacturing products precisely in line with specifications
becomes crucial to meet customer requirements effectively.

Scenario 2: Car Wider Than Garage

- Representation: The car is wider than the garage.

- Analogy: Even if the wider car is parked precisely, it will not fit within the specified bound-
aries.

- Interpretation: Similarly, in a process where product variability exceeds specification limits,

meeting the exact midpoint becomes inconsequential as the products will not fit within the
required specifications.

57
Scenario 3: Car Significantly Smaller Than Garage

- Representation: The car is significantly smaller than the garage.

- Analogy: When the car is much smaller, precision in parking, such as centering, becomes
inconsequential. The car easily fits within the garage with ample space on either side.

- Interpretation: In a process with low variability where products comfortably fit within the
specification limits, the precision of meeting the midpoint becomes less critical as the products
consistently meet the specified requirements.

This analogy underscores the significance of process control and minimal variation in consistently
aligning with specifications and meeting customer expectations.

Example:

A manufacturing process produces a certain type of bolt with a mean diameter of 2 inches and
a standard deviation of 0.05 inches. The lower and upper specification limits are 1.90 and 2.05
inches, respectively. Calculate the process capability and determine if the manufacturing process is
capable of meeting the specification requirements.

Solution:

Given parameters:
Mean Diameter (µ) = 2 inches

Standard Deviation (σ) = 0.05 inches

Lower Specification Limit (LSL) = 1.90 inches

Upper Specification Limit (USL) = 2.05 inches

Process capability is given by 6σ, where:

6σ = 6 × 0.05 = 0.30

Tolerance = USL − LSL = 2.05 − 1.90 = 0.15

Since process capability is greater than the tolerance (0.30 > 0.15), the products do not meet the
desired specifications.

58