Page

Statistical
Process Control
(SPC)
 Page 2

Table of Contents

Introduction to Continuous Improvement and Statistical Process Control ........ Module 1

Basic Statistics for Statistical Process Control ............................................... Module 2

Sources of Variations: Common and Special Causes ....................................... Module 3

Distribution and Statistics for Normally Distributed Data................................. Module 4

Evaluation of Capability for a Stable Process ........ ....................................... Module 5

Average and Range Charts ............................................................................ Module 6

Interpretation of Control Charts ..................................................................... Module 7

Strategies for Sampling Data ........................................................................ Module 8

Average and Standard Deviation Charts ........................................................ Module 9

.
Median and Range Charts ............................................................................. Module 10

Individual and Moving Range Chart ...................... ....................................... Module 11

Proportion and Number of Nonconforming Charts ...................................... Module 12

Number of Nonconformities and Nonconformities per unit Charts ............. Module 13

Application Guidelines ................................................................................. Module 14

Application Examples .................................................................................. Module 15

Appendices ................................................................................................... Module 16

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 3

Short List of Notations

x = a number (data, sample reading).

x = the average (mean) of a set of x.

x = the average of the averages x , also called grand average.

≈
X = the median of the numbers x. (Symbol ≈ used for median)

≈
X = the average of the medians

∑X = the sum of numbers x

n = the total number of data

K = the number of samples (if 100 parts are grouped into 20 groups of 5 each for
measurement purposes, then K = 20, n = 5 for group, n = 100 for all data.)
M = the mode of numbers
S = the standard deviation of numbers
( SX , S X , S R , etc.).
S = the average of the Standard deviation (S)
R = the range of numbers

R = the average of the ranges (R).

µX = The average of all possible number (even of parts that are not
included in measurement)
σ X = the standard deviation of all possible numbers

< = less than

< = less than or equal to

> = greater than

> = greater than or equal to

= square root of quantity under the radical sign

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 4

Course Overview

Industrial production often involves mass production of the same part. For such parts to properly
assemble and function in the final product, it is necessary to keep the variation in quality
characteristic to a minimum. The variation in quality characteristics are caused mainly by two
sources; known as common and special causes. Statistical process control (SPC) is used to study
the process performance and understand sources of variation with the intention of making
corrective actions to reduce variation.

This brief session covers the basic concepts of statistical analysis and their application to
practical problems in process control. It will deal with such standard tools as histograms, X-bar
and R charts, process capability studies and sampling plans. It is intended to help attendees
decide if SPC will be helpful their activities and whether further training should be sought before
applications. Attendees to this session learn how to use histograms, Pareto charts, scatter
diagrams, etc. The primary focus of the course will be to help attendees develop a working level
understanding of the normal distribution and how to construct control charts for variable and
attribute data. The attendees are expected to gather understanding of how to use DOE results for
SPC, calculate process capabilities, and learn how to communicate with design engineering in
terms of process capabilities.

Instructor
Ranjit K. Roy, Ph.D., P.E. (M.E.) is an engineering consultant specializing in
Taguchi approach of quality improvement. Dr. Roy has achieved international
recognition as a consultant and trainer for his down-to-earth teaching style of the
Taguchi experimental design technique, project management, and several other
quality engineering topics. Dr. Roy began his career as senior design engineer with
Burroughs Corporation following completion of graduate studies in engineering at the University
of Missouri-Rolla in 1972. He then worked for General Motors Corp. (1976-1987) assuming
various engineering responsibilities with his last position as that of reliability manager. Dr. Roy
is a fellow of the American Society of Quality.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 5

SPC STEP 6
Learning Steps Learn the rules of
out-of-control
detection based on
STEP 5 probability of
Understand the logic occurrence.
behind Examine control
standardization of charts carefully
data collection and
simplification of
STEP 4 formula for
Learn what are: calculations of limits
- In control for control charts.
- Stable
- In specification
- Capable

STEP 3
Understand that
process variation is the
result of COMMON-
CAUSE and
SPECIAL-CAUSE
variations
(Fundamental
Equation of SPC)

STEP 2
Learn the Normal Theory
and the Central Limit
Theorem and their
implication well.

- if X is normal the
distribution of the process
STEP 1 average is also normal.
Understand
- If distribution of X is not
properties of
normal, the distribution of
normal
sample mean is still
distribution.
approximatly normal.
- All data to fall
within +/- 3 Std.
Dev. under
common cause
variations

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 6

SPC
Application Steps
(Observe the Process - Define Standard -Compare Performance)

Carefully examine
data for out-of-
control situations.
Mark points based
on the “Rules”.
3
Calculate limits
and construct the
desired control
charts.

Plot data.

Collect data and

calculate sample
averages, ranges,
grand average,
1 average range, etc.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 7

What is
Statistical Process Control ?

Control
Process What does it mean?

What does it mean? To correct:

Statistical Any task that affects
- Correct to what?
- How do would we
cost and performance know that it is not
What does it mean?
of products or services: right?
- Incoming order - How bad are we?
Inexact:
processing - How far off is it from
It is never the exact
- Payroll checks what it should be?
value.
- Accounts payable - How fast do we need
- Part fabrication to correct the situation?
Predicted value based
(milling, turning,
on more than one
welding, molding, etc)
observed data: To keep it where it
- Typing/data entry
Data used for should be:
- Assembly of parts
calculating any statistic - How do we find out
- Soldering
(numeric value) are where should it be?
more than one. - What can we do to
Any operation or
sequence of operation bring it where it should
Estimate based on a be?
with measurable
small data sample from - How do we know it is
output:
the whole: not where it should be?

Valid for average

performance:

Conclusions based on
past observations:

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 8

Module 1
Introduction to
Continuous Improvement and
Statistical Process Control

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 9

Introduction to Continuous Improvement and Statistical Process Control

What is Quality?

- Meeting Specifications

- Preventing Defects

- Practicing Continuous Improvement

“Quality is meeting or exceeding customer expectations at all times in regards to performance,

durability, reliability, and delivery at a cost that represents value.”

“Quality is producing a reliable & durable product to exact standard at the lowest cost for the
customers.”

How?
- Product Control
- Process Control

In 1970’s W. Deming, J. Juran and others emphasized the need to shift focus from product
control to process control.

Stop depending on inspection to achieve quality. Eliminate the need for

extensive inspection programs by designing and building quality into the
product or service in the first place.

- W. Edward Deming
A Sample Process
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 10

Transformation Machine * Output

* Input
* People * Products
- Work process * Services
* Machine - Add value to input
* Materials * Customer
- Assemble materials satisfaction

Product Control Vs. Process Control

Items/Activities How it is done

Product Control Process Control

Practical Objectives Keep Product Characteristics Try to be on target with

within specifications minimum variation satisfying
the quality standard

Ideal Objectives Detect and eliminate non- Assure that non-conforming

conforming products products are not produced

Improvement Strategy Inspect and maintain outgoing Institute and maintain

product quality continuous quality improvement
procedures

Statistical Techniques Acceptance Sampling Control Charts (SPC)

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 11

Module 2
Basic Statistics for
Statistical Process Control

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 12

Types of Data (Measured Data)

• Variable Data
- Diameter (mm)
- Length (inches)
- Weight (pounds)
- Torque (Inch-Pound)
- Tension (pound)
- Hardness, etc.

• Attribute Data
- Go/No go (% of no go)
- Conforming/Non-conforming (number non-conforming)

Example: Defects found by inspecting 1000 parts.

Day 1 X1 = 50 Day 6 X6 = 70 Day 11 X11 = 90

Day 2 X2 = 60 Day 7 X7 = 60 Day 12 X12 = 30
Day 3 X3 = 48 Day 8 X8 = 90 Day 13 X13 = 42
Day 4 X4 = 20 Day 9 X9 = 80 Day 14 X14 = 60
Day 5 X5 = 71 Day 10 X10 = 60 Day 15 X15 = 100

Data plot in a number scale

0 10 20 30 40 50 60 70 80 90 100 110

Histogram
Shape of the Histogram - Distribution

Let us use the following data to define the tools that measure a distribution.

Example: Sample readings of a few used batteries were as follows:

Battery no. 1 X1 = 6 Volts

Battery no. 2 X2 = 3 Volts
Battery no. 3 X3 = 5 Volts
Battery no. 4 X4 = 4 Volts
Battery no. 5 X5 = 9 Volts
Battery no. 6 X6 = 6 Volts
Battery no. 7 X7 = 11 Volts
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 13

Notice the notations used to indicate a specific data. For example, the third data point or the third
observation, 5, is described using the notation

X3 = 5

Notation is useful in explaining methods used and communicating ideas without actually using a
specific number. However, most often we write data set without using notations like
in this case the data set is:

6 3 5 4 9 6 11

Shape of Data

The shape of the data is the shape of its distribution and is obtained by plotting the data points in
a histogram, which can be drawn by placing one ball for each data point on the corresponding
value along the x-axis.

2 3 4 5 6 7 8 9 10 11

Shapes of Distributions

Normal Distribution (ND - Bell shaped curve)

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 14

- There are a large number of situations that exhibit ND

- In absence of any special causes many data will show ND

- A vast majority of industrial product and process data follow ND

Example: Human heart beat, Diameter of a machined shaft, Voltage of a particular brand of
battery, etc.

Predicting Shapes of Distributions ...

* Weights of apples in a bad year

* Grade point averages of freshman year in reputed schools
* Ages of people attending meetings of retirement planning

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 15

Statistical Tools for analysis of data sets

There are a number of statistical tools are used for extracting information hidden in a set of
numbers.

Questions which may arise while reviewing a Normal Distribution

Where is the stem of the “umbrella”?

How wide is it?
What shape is it?

Statistically speaking, how do we measure:

• Central Tendency or location of the center

- Mean or Average (*)
- Median
- Mode

• Variation or Spread or closeness of data to the center

- Range (*)
- Variance
- Standard Deviation (*)

• The Shape
- Symmetry
- Skewness
- Flatness
- Similarity to normal distribution (*)

(*) More popular in SPC studies

1. Location of the center -

measured by Average ( x ), Median (x), and Mode

Average, Mean or Balance point of data

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 16

Mean of the data is obtained by simply adding all data and dividing by the total number of data.

_
Mean or Average X = ( 6 + 3 + 5 + 4 + 9 + 6 + 11 ) / 7 = 6.285

Median

The median of the data is the observation, which is in the middle of the data set when the entire
observation is placed in ascending order, i.e., from smallest to the largest.

data: 3 4 5 6 6 9 11

The observation, 6, which is in the 4th place of the total of 7 observations, is the median of the
data.

x% = 6 (notation uses curl over the character X)

When there are even number of observations, the median will be the average of the two closest
nubers.

For a set of data: 9 10 12 5 19 20, median is the number between 12 and 15.

x% = (12 + 15)/2 = 13.5

Mode

Mode is the number that occurs most frequently in a set of numbers. In this case the number 6
occurs twice, where as the other numbers occur only once. The mode in this case is 6.

Exercise:

Data: 5 7 2 7 3 9

Find:

Average = Ans: 5.5

Median = Ans: (5+7)/2 = 6

Mode = Ans: 7

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 17

2. Closeness of data to the center -

measured by Range (R), Variance ( S2 ), and Standard Deviation ( S )

Range, Variance, and Std. Deviation

data: 2 5 3 7 4

Range

R = Difference between the largest and smallest data

= 7-2 = 5

Variance

Sum of Squares of distance from the average

S2 =
Number of observations minus 1

average = (2 + 5 + 3 + 7 + 4)/5 = 4.2

S2 = [(2 - 4.2)2 + (5 - 4.2)2 + (3 - 4.2)2 + (7 - 4.2)2 + 4 - 4.2)2]/(5 - 1)

= (4.84 + 10.24 + 1.44 + 10.24 + 1.44) /(5-1)

= 7.05

Standard Deviation(s) = variance

S = S 2 = 7. 05 = 2. 655

3. Shape of the curve -

measured by Symmetry, Skewness, Flatness, and Similarity to ND

In SPC studies data is assumed to be normally distributed.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 18

Subgroup, Sample, and Population

Sample Subgroup
Population

Sample - a selected set of data from the total number of possible data. Consider a milling
machine producing a total of 60,000 parts. If only 20 sets of 5 parts (5 x 20 = 100) out of the
total are inspected, then:

Sample size = 100,

Subgroup size = 5, subgroups are smaller sample (20 subgroups)

Population is 60,000.

Generally for statistical calculations, the sample size is subdivided into smaller groups. The
subgroup size of 5 or 10 data are common. With 5 data in a subgroup, 100 data sample is
divided into 20 subgroups.

Subgroup, Sample and Population Statistics

Data:
Subgroup 1: 12 14 11 16 12 Avg = 13

Subgroup 2: 12 14 12 15 16 Avg = 13.8

Subgroup 3: 11 16 13 15 12 Avg = 13.4

Subgroup averages

x1 = 13, x 2 = 13. 8, x 3 = 13. 4

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 19

Sample average x = (13 +13.8 + 13.4)/3 = 13.4

(Grand average)

Also, the average of all data, that is

(12 + 14 + 11 + 16 + 12 + 12 + 14 + 12 + 15 + 16 +11 + 16 + 13 +15 +12)/15

= 201/15 = 13.4 (same as the average of all averages)

Subgroup Ranges

R1 = 16 - 11 = 5
R2 = 16 - 13 = 4
R3 = 16 - 11 = 5

Sample Range = (R1 + R2 + R3)/3 = (5 + 4 + 5)/3 = 4.666

(average range)

Likewise standard deviation for the sample can be calculated.

Practice Problems (Fill in the missing answers)

1. Data: 7 4 3 8 6 Avg. = 5.599 Std. Dev.= 2.073 Range =

2. Data: 12 10 11 9 8 Avg. = 10 Std. Dev.= Range = 4

3. Data: 2 4 3 5 5 Avg. = 3.8 Std. Dev.= Range = 3

4. Data: 14 18 15 12 23 Avg. = Std. Dev.=4.277 Range =

5. Data: 9 11 13 14 12 Avg. = Std. Dev.= 1.923 Range =

6. Data: 6 9 12 11 12 Avg. = 10 Std. Dev.= Range = 6

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 20

Module 3
Sources of Variations:
Common and Special Causes

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 21

Sources of Variations: Common and Special Causes

Variation is natural and is expected in every process or production output.

Sources of variations

• People
- Operator fatigues
- Operator experience on the job
- Training

• Equipment
- Setup
- Old/New machine
- Maintenance

• Production method
- Rate of production
- Inspection & downtime

• Materials
- Uniformity
- Change of property

• Environment
- Temperature
- Humidity
- Dust in the air

Depending on the source, the causes of variations are classified into two groups.

• Common-cause variation
- Difficult to reduce by ordinary approach of fine-tuning. It usually require significant
changes to the process
- The primary responsibility in reducing this variation rests with the management

• Special-cause variation
- Often possible to reduce by making adjustment at the point of production
- Process operator and immediate management usually have the primary responsibility

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 22

Fundamental Equation of Statistical Process Control

Total variation of Special-cause Common-cause

production output Variation Variation

One of the objectives of SPC is to reduce variation in production output.

- Identify common-cause variation and special-cause variation

- Remove special-cause variation
- Reduce common-cause variation

“I should estimate that in my experience most troubles and most

possibilities for improvement add up to proportions something like
this:

• 94% are common-cause: they belong to the system (responsibility

of management)
• 6% are special-cause variation (responsibility of operators) ”

Control chart (to be studied later)

A control chart shows variation in a production process and it contains information, which can be
used to detect the presence of special-cause variation.

The control chart for a process utilizes information from the process when it is only subjected to
common-cause variation. When actual performance is monitored and data plotted in the control
chart, information about the possible presence of special-cause variation in the process is
revealed.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 23

Process stability

- A stable process has only common-cause variation and no special-cause variation.

- Most processes that are stable show distributions that are normal or close to being normal.

- If a process is stable and shows normal distribution for individual values of the output, or for
the subgroup averages and ranges, the process is considered to be in control

- Special-cause variation causes instability in the process and must be corrected before the SPC
data is collected

In Control

A process is in control if over time there is no change in the distribution of the process
output i.e., there is no change in the process center and there is no change in the process
variation.

Mean/Average Variation

No change No change In control and Stable

Change No Change Out of control and Unstable
No Change Change ‘’
Change Change ‘’

In Specification

A process is considered in specifications when the distribution is within the lower (LCL) and
upper (UCL) control limits.

A stable process can be:

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 24

• In control and in spec.

• In control but not in spec.

• Out of control but in spec.

• Out of control and out of spec.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 25

Process capability (to be studied later)

A process can be in control but not in spec but capable.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 26

Module 4
Statistics for
Normally Distributed Data

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 27

Statistics for Normally Distributed Data

Normal Distribution (ND)

Distribution - refers to the shape of the histogram created from a set of numbers (data).

Data - diameter of a shaft, ages of people, delivery times, number of faults, etc.

Histogram - a bar graph made using the frequency of occurrence of data.

Frequency - a number that represents how many times a particular number occurs in a data set.
In a data set if number 5 occurs three times then frequency of occurrence of number 5 is 3.

Normal - a natural or usual behavior.

Data collected from process under normal operating condition will follow a specific type of
distribution. Mathematicians using different mathematical equations have modeled a few
common distribution shapes: Normal, Binomial, Exponential, etc. The data appear to follow
close similarities with normal distribution (ND).

Characteristics of ND:
It looks like a bell.
It is symmetrical about the center.
Ends are almost touching the x-axis (never actually touching though).
Highest point is at the center.

Practice drawing ND:

Draw ND with mean at 5.

Draw a narrower ND at 5.
Draw a flat ND at 7.

Where does the ND shape come from?

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 28

The shape of the normal distribution curve (word curve is generally dropped) comes from the
shape of the HISTOGRAM created from a type of data that fits a normal distribution. The shape
can be easily seen by drawing the histogram.

Histogram - a bar graph of the frequency (number of times a particular data occurs) of the data
set. Consider a data set:

5 7 8 3 2 6 5 8 9 7 and 6

This data set has

One(1) 2 frequency of occurrence of number 2 is 1

One(1) 3 frequency of occurrence is 1
Two(2) 5 frequency of occurrence is 2
Two(2) 6 etc.
Two (2) 7
Two (2) 8
and One (1) 9

If we were to plot the frequencies in a bar graph, it will look like the following graph.

For Plotting
purposes data may be grouped in to fixed ranges.

Data:
6.23 6.35 6.69 6.85 6.99 6.11 6.42 6.61 6.8 6.96 6.45 6.52 6.9 6.91
6.5 6.54 6.76 6.38 6.56 6.72 6.65 6.60 6.53 6.52

The data is grouped in ranges between 6.1 -6.3, 6.31 - 6.5, 6.51 -6.7, etc.
There are 2 data that falls in the first group, i.e., between 6.1 - 6.3, etc.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 29

Depending on the nature of the data, the histogram might exhibit many other shapes.

But the most common data are found to exhibit normal distribution.

More facts about normal distribution:

Research has determined that the ND is centered about the average and the entire area under
the curve is covered less than 3 times the standard deviation from the center.

1 ( x−µ )2
f ( x) = e 2σ 2
s 2∏

-3σ -2σ -1σ 1σ 2σ 3σ

68.27 %
95.45
99.73
Area under Normal Distribution

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 30

Percentage of Total under Normal Distribution Curve

Within 1 Std. Dev. from Average 68.268949%

2 “ “ 95.449974%
3 “ “ 99.730020%
4 “ “ 99.993666%

Area under the “tails” of the ND:

at 2 St. Deviantion: 100-95.449974 = 4.55003%

What if Distribution is not normal?

Not all data will show normal distribution and there is no guarantee the data collected will follow
a normal curve.

However, whether the collected data (from product, process or service characteristic) is normal
or not, the sample averages x fall in a normal distribution. (Central Limit Theorem).

Normal Theory: If the distribution of the measurement, X is normal then the distribution of x
is normal.
.

Central Limit Theorem: whether x is normal or not, if sample size n (number of subgroups)
is large, x is still normal.

Facts:
µx = µx
Average of sample averages is equal to the average of all data (population).

σx
σx = S X , S X are estimates of standard deviations
n
Standard deviation of normal distribution plot of sample averages is
equal to standard deviation of population (all data) divided by square
root of sample size.

If the distribution is normal, then 99.73% of x will be within µ ± 3σ . This means that if
limits are set at 3 standard deviations apart from the mean, only .27% of the observations. In
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 31

other words, no more than 3 (2.7) observations out of a thousand are expected to be outside the
limits.

Data and data reduced to any other form such as average, range, standard deviation, etc. all can
be plotted and distributions obtained. When the distributions are normal, its characteristics can be
used to identify normal behavior.

Example:

data: 2 5 3 7 4

Range = Difference between the largest and smallest data

= 7-2 = 5

Sum of Squares of distance from average

Variance = S2 =
(Number of observations - 1)

average = (2 + 5 + 3 + 7 + 4)/5 = 4.2

S2 = [(2 - 4.2)2 + (5 - 4.2)2 + (3 - 4.2)2 + (7 - 4.2)2 + 4 - 4.2)2]/(5 - 1)

= (4.84 + 10.24 + 1.44 + 10.24 + 1.44) / 4

= 7.05

Standard Deviation(s) = 7.05 = 2.655

Sample, Subgroup, and Population

Sample - a selected set of data from the total number of possible data. Consider a milling
machine producing a total of 60,000 parts. If only 20 sets of 5 parts (5 x 20 = 100) out of the
total are inspected, then:
Sample size = 100, and
subgroup size = 20

Subgroups are smaller sample

Population is 60,000.

Generally for statistical calculations, the sample size is subdivided into smaller groups. The
subgroup size of 5 or 10 data are common. With 5 data in a subgroup, 100 data sample is
divided into 20 subgroups.

Subgroup, Sample and Population Statistics

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 32

Data:
Subgroup 1: 12 14 11 16 12 Avg. = 13
Subgroup 2: 12 14 12 15 16 Avg. = 13.8
Subgroup 3: 11 16 13 15 12 Avg. = 13.4

Sample average X = (13 + 13. 8 + 13.4 )/ 3 = 13.4

(Grand average)

Subgroup Ranges

R1 = 16 - 11 = 5
R2 = 16 - 13 = 4
R3 = 16 - 11 = 5

Sample Range R = (R1 + R2 + R3)/3 = (5 + 4 + 5)/3 = 4.666

(average range)

Likewise standard deviation for sample can be calculated.

The Standard deviation of sample averages (three averages)

S X = [( 13 - 13.4 )2 + (13.8 - 13.4 )2 + (13.4 - 13.4)2 / (3-1) ] 0.5

= (0.16 )0.5 = 0.40

The standard deviation for the entire data (15 numbers) will require the use of all 15 numbers.
This process is generally complicated and not done. Instead, the standard deviation is estimated
from the value of average range ( R ).

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 33

Module 5
Evaluation of Capability
for a Stable Process

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 34

Process Capability

Process capability describes the relationship between actual production output and acceptable
production output. (Actual production output is compared with the specification limits)

Process Control

Control charts are used to compare values from a sample production output relates to control
limits. (Sample data are compared with the control limits)

Common Terms

• Upper specification limit (USL) the maximum allowable measurement of a product or service
characteristics
• Lower specification limit (LSL) the minimum allowable measurement of a product or service
characteristics
• Target value - value of the measured value of the product or service characteristic that the
customer ideally wants at all time

Process Capability Analysis

The distribution of process output to the specification limits is done when:

- the process is in statistical control

- it is a single product or service characteristic
- capability of the process is dependent on the distribution of characteristic measured
- capability of the process is related to the target value and the percentage of production
parts outside the specification limits

Desirable Characteristics of Measure of Capability (Indices)

• Should be simple
• Pure number
• Relate to measured characteristic

There are two common indices used.

Cp - The process potential index

It represents the ability of the process to satisfy engineering or manufacturing specification.
USL - LSL
Cp =
6S X

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 35

Cp does not depend on the location of the center of output and the target.

Cpk - The Process Capability Index

It indicates the actual performance of the process relative to engineering or manufacturing

specification.

X - CSL
C pk =
3S X

depends both on the specification limits and the target.

where the estimated Standard deviation from the population (Sx),

R
S =
x d
2
and
CSL = USL or LSL depending on whichever is closer to the grand average

X= µ x (The mean of sample mean is always the same as the population mean)

Cp = (Difference between the specification limits )/ 6σ

Cpk = (Population Mean - Nearest specification limit )/ 3σ

Note: Cp is what Cpk can be.

Cpk = 1 represents 3σ quality

Cpk = 1.33 represents 4σ quality
Cpk = 1.66 represents 5σ quality
Cpk = 2.00 represents 6σ quality

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 36

Cp =

Cpk =

LSL Target USL

SL

Cp =

Cpk =

LSL Target USL

SL

Cp =

Cpk =

LSL Target USL

SL
Cp is the same regardless of the mean performance. Cpk accounts for product center(preferred
index). It can be viewed as a distance measurement of the product distribution mean from the
nearest specification limit in terms of a “measuring yardstick” which is three standard deviation
long.

Cpk - Characteristics

• Increases as the mean of the product output fall close to the target
• Decreases as the variation(std. Deviation) of the product performance increases

Product and Process Capabilities

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 37

Product Capability ∝ 1 /σ σT
(when Cp and Cpk are calculated using total variation)

Process Capability ∝ 1 /σ σ CC
(when Cp and Cpk are calculated using only common cause variation)

σT 2 = σCC 2
+ σSC 2

When the process is in control and all variation is due to common cause, then

σSC = 0 and σT 2 = σCC 2

Product and Process Capability Relations

Product Capability
Cpk

In Control On target

Product Potential Cp
Process Capability Cpk

On target In Control

Process Potential Cp

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 38

Review of Capability Indices

Figure 5.1

LSL Target USL

Cp = 1 Cpk = 1
Notes: When distribution is centered both indices have the same value.
USL - LSL = 12 units. Std. Dev. = 2 units. Cpk = 6/6 = 1

Figure 5.2

LSL Target USL

Cp = 2 Cpk = 1
Notes: When distribution is not centered, Cpk is smaller than Cp.
USL - LSL = 12 units. Std. Dev = 1 units. Cpk = 3/3 = 1

Figure 5.3

LSL Target USL

Cp = 2 Cpk = 1

Notes: When distribution is not centered Cpk is smaller than Cp

regardless of the center location.
R. Roy/Nutek, Inc. USL - LSL
Statistical = 12
Process units.
Control Std. Dev.
– Management = 1 units.Version 0509
Overview
 Page 39

Figure 5.4

LSL Target USL

Cp = 6 Cpk = 1
Notes: Extremely off-centered distribution for the same Cpk value.
USL - LSL = 12 units. Std. Dev. = 1/3 units. Cpk = 1/1 = 1

Figure 5.5

LSL Target USL

Cp = 6 Cpk = 1

Notes: Extremely off-centered distribution, close to the LSL.

USL - LSL = 12 units. Std. Dev. = 1/6 units. Cpk = 0.50 / .050 = 1

Figure 5.6

LSL Target USL

Cp = 1 (Shallow) Cpk = 1 Cp = 2 Cpk = 2

Notes: Two centered distributions, one is twice as good as the other.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 40

Figure 5.7

LSL Target USL

Cp = 1 (Shallower) Cpk = 1 Cp =6 (Narrower) Cpk = 6

Notes: Centered distribution of different shapes.

Figure 5.8

LSL Target USL

Cp = 2 Cpk = 1 Cp = 6 Cpk = 2

Notes: Two off-centered distributions of different shapes.

Figure 5.9

LSL Target USL

Cp = Cpk =

Notes:
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 41

Figure 5.10

LSL Target USL

Cp = Cpk =

Notes:

Figure 5.11

LSL Target USL

Cp = Cpk =

Notes:

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 42

Dr. Genechi Taguchi proposed a modified Capability Index (Cpm) which makes use of the
average deviation of data (X) from the target (X0) to compute the standard deviation (σT).

C pm =
USL - LSL
where σ =
∑ ( X - X 0 )2
6σ T T
N

(N - 1)
also σT = σ c2 + (X 0 - X) 2
N

Notice that σT , is same as the Mean Squared Deviation (MSD) defined for the Nominal is the
Best quality characteristics by Taguchi.

Yet another index proposed by Dr. Carol Vale, Ford Motor Company, makes use of the data
spread to compute the capability index (Cpa). In this formulation, the distance (XC) of the data
group (histogram bar) from the nearest specification limit is divided by the distance of the
specification limit from the target.

Target - LSL
C pa =
Xc

Figure 5.12

LSL Target USL

Cp = >1 Cpa = 1.5/6 =

0.25
Notes:

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 43

Control chart and the resulting distribution.

USL USL

UCL

LC
LU
LSL LSL

Distribution after continuous improvement.

USL USL

UCL

LCL
USL
LSL LSL
USL

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 44

Module 6
Average and Range Charts

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 45

What are Average ( X ) and Range (R) Charts?

These charts are plots of Average ( X ) and Range(R) over time with two additional lines drawn
on the plots. the two additional lines are known as the control limits.

Average and Range are typically done for measurement characteristics of type

• Thickness ( cm, inches, etc)

• Length (feet, mm, inches, etc)
• Diameter (inches, mm, etc)
• Tensile strength (pounds per sq. inch, kg per sq. cm. etc.)
• Time (minutes, seconds, etc)
• Number of defects (in units)

How to Construct Average ( X ) Chart

1. Calculate X and R for all samples

2. Calculate average of the sample averages ( X ) and the sample Range average ( R )

3. Calculate control limits

U / LCL = X+ 3S
X X

If the process is in control, 99.73% of the sample averages will fall within the control limits.

S = “best estimate” of Standard Deviation of the sample averages( X )

X
S = “best estimate” of the measured characteristic (X )
X

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 46

Estimate of the Standard Deviation (of Population, X)

σx = The standard deviation of population of the measuremen ts of the process characteri zes
(can be calculated by directly using all the readings)

S, S = Sample Standard deviation and average of the sample standard deviation.

R, R = Sample range and average of the sample ranges

σ x can be estimated as (S x )

R
using R as Sx =
d2

S
using S as Sx =
C4

where constants d2 = 2.326 and C4 = .9400 for subgroup/sample size = 5

Estimate of the Standard Deviation (of Sample Means)

SX = Standard deviation of sample means ( X )

Sx = Standard deviation of sample population (x)
n = number of measurements in each sample
X, X = sample mean and mean of the sample means

The standard deviation of the sample means is always smaller than the population mean and is
inversely proportional to the square of the number of samples.

SX
SX =
n
Thus, from the definition of population standard deviation, SX, above,

R
SX = using R
d2 n

S
SX = using S
C4 n

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 47

Control limits for Sample Means

Limits =
UCL X = X + 3 S X

UCL X = X + A 2 R
or
LCL X = X − A 2 R

where
3 3
A2 = = = 0.577
d2 n 2.326 5

(for sample size n = 5, d2 = 2.326)

Control Limits for Sample Range (R)

UCL R = D 4 R = 2114
. R

LCL R = 0

(for sample size n = 5, and D4 = 2.114)

Common Plots of X-bar and R Charts

P lo t s a m p le X o n X c o n tro l c h a rt
P lo t s a m p le R o n R c o n tr o l c h a r t

How to analyze data plots on the range charts

i) Points beyond the control limits

The occurrence of one or more points beyond either control limits primary evidence of non-
control at that point. If the process is in a state of control and only variation due to common
causes is present, any points beyond control limits would be rare.

A point above upper control limit for ranges is generally due to one of the following:
• The control limits calculated erroneously
• The piece-to-piece variability or the spread of the distribution has worsened, either at that
point or as part of the trend
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 48

• The measurement system has changed.

• The measurement system lacks appropriate discrimination

A point below the lower control limit (for n > 6) is generally due to one of the following:
• The control limit calculated is wrong
• The spread of the distribution has decreased
• The measurement system has changed

Patterns or Trends Within the Control Limits:

The presence of unusual pattern or trends even when all ranges are within control limits, can be
evidence of non-control. Unfavorable trends are subject of investigation and correction.
Favorable trend + should be studied for possible permanent improvement of the process.

ii) Runs – The following are signs of process shift

• 7 points in a row on one side of the average
• 7 points in a row that are consistently increasing or decreasing

Possible causes:
-- change in measurement system
-- greater range indicates trouble and should be investigated
-- reduced range should be studied for permanent process change.

iii) Obvious nonrandom pattern -- presence of distinctive pattern offers clues to special
causes.

Distance from R - for a process in a state of control, 2/3 of the data should be within 1/3 of
the control limits and 1/3 of the data should be within the outer two thirds of the region.

If substantially more than 2/3 of the points lie close to the process average (for 25 subgroups if
90% are in the middle third of the control limits), look for the following causes.

• Control limits miscalculated

• Data have been edited
• Subgroups have been stratified and have different averages.

If significantly fewer than 2/3 of the data points lie close to the process average (40% or fewer),
suspect one of the following.

• Control limits have been miscalculated.

• The sampling method used caused subgroups to contain measurements from two or more
very different process streams. If several process streams are present they should be
identified and tracked separately.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 49

Example 6.1 Data Type: Seat Deformation

Sample X1 X2 X3 X4 X5 Average Range Std.Dev.

1 124 122 129 118 122 123.00 11 4.00

2 126 131 126 122 125 126.00 9 3.24
3 119 127 124 128 132 126.00 13 4.85
4 122 120 121 117 125 121.00 8 2.92
5 119 127 122 132 124 124.80 13 4.97
6 118 128 129 126 130 126.20 12 4.82
7 117 130 119 136 114 123.20 22 9.36
8 113 130 122 123 116 120.80 17 6.61
9 116 118 112 128 125 119.80 16 6.57
10 126 119 123 129 129 125.20 10 4.27
11 123 133 129 115 127 125.40 18 6.84
12 128 106 122 121 120 119.40 22 8.11
13 141 130 129 122 129 130.20 19 6.83
14 120 121 122 135 116 122.80 19 7.19
15 127 121 116 118 115 119.40 12 4.83
16 122 136 126 124 139 129.40 17 7.60
17 119 122 123 120 124 121.60 5 2.07
18 117 125 137 126 133 127.60 20 7.73
19 128 122 129 113 129 124.20 16 6.91
20 129 124 117 119 121 122.00 12 4.69

Grand Avg.= 123.900 Avg.Range = 14.550 Avg.Std.Dev.= 5.721

Control limit for X-bar chart

UCL X = X + A 2 R (Upper limit) = 123.9 + .577 x 14.55 = 132.29
LCLX = X − A 2 R (Lower limit) = 123.9 - .577 x 14.55 = 115.50
(Constant A2 = 0.577 for sample size = 5) One Std.Dev = .577x14.55/3 = 2.80

Calculation of C P & C PK (assuming USL = 150, LSL = 110, Target = 125)

SX = R d 2 (estimate of σX ). = 14.55 / 2.326 = 6.25 (d2 =2.326)

X = µX (estimate of Avg.). = 123.90

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 50

USL − LSL
CP = = (150 -110 ) / (6 x 6.25) = 1.066
6S X
X − CSL
C PK = = (123.9 -110 ) / (3 x 6.25 ) = 0.741
3S X

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 51

Plant Dept. Operation Eng. Specs. Part#

Machine# Date: Measure Sample size Part Name

X-bar Chart (Example 6.1) LCL = 115.50 x = 123.9 UCL = 132.29

135

132

130

125
123.

120

115

110
1 5 10 15 20

Measured Data (Units of mesurement: )

Date Time 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Rea 1
di 2
n 3
g 4
s 5

x
S

LCL = x = UCL =

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 52

Example 6.2 Data Type: Pressure Plate Height

Sample X1 X2 X3 X4 X5 Average Range Std.Dev.

1 5.74 5.80 5.72 5.78 5.63 5.73 0.17 0.07

2 6.17 6.29 5.69 6.14 6.14 6.09 0.60 0.23
3 7.09 6.63 5.65 6.39 6.35 6.42 1.44 0.52
4 5.80 6.15 5.96 5.97 5.48 5.87 0.67 0.25
5 6.85 5.55 6.16 6.25 5.43 6.05 1.42 0.58
6 6.19 6.29 6.37 5.84 5.36 6.01 1.01 0.42
7 5.97 6.13 5.25 5.80 5.45 5.72 0.88 0.36
8 5.51 5.71 7.09 6.19 5.70 6.04 1.58 0.64
9 5.81 5.92 5.26 5.89 5.85 5.75 0.66 0.27
10 6.72 5.70 5.54 5.99 5.76 5.94 1.18 0.46
11 6.02 5.87 5.67 5.27 6.17 5.80 0.90 0.35
12 5.80 6.08 6.51 5.83 6.17 6.08 0.71 0.29
13 5.84 5.25 5.43 5.51 5.50 5.51 0.59 0.21
14 6.33 5.68 5.49 5.14 6.27 5.78 1.19 0.51
15 6.46 6.02 6.37 7.08 6.04 6.39 1.06 0.43
16 5.27 6.14 5.74 5.71 6.25 5.82 0.98 0.39
17 5.78 5.84 5.61 6.50 5.50 5.85 1.00 0.39
18 5.78 6.20 6.09 6.27 5.46 5.96 0.81 0.34
19 5.73 4.80 6.37 6.61 5.75 5.85 1.81 0.70
20 5.61 6.74 5.57 5.81 5.80 5.91 1.17 0.48

Grand Avg.= 5.928 Avg.Range = 0.991 Avg.Std.Dev.= 0.395

Control limit for X-bar chart

UCL X = X + A 2 R (Upper limit) = 5.928 + .577 x .991 = > 6.5
LCLX = X − A 2 R (Lower limit) = 5.928 - .577 x .991 = > 5.3
(Constant A2 = 0.577 for sample size = 5) One Std.Dev = .577x0.991/3 = > 0.20

Calculation of C P & C PK (assuming USL = 7.5 , LSL = 3.5, Target = 5.5)

SX = R d 2 (estimate of σX ). = .991 / 2.326 =0.426 (d2 =2.326)

X = µX (estimate of Avg.). = 5.928

USL − LSL
CP = = (7.5 - 3.5 )/(6 x 0.426) = 1.565
6S X
X − CSL
C PK = = (7.5 - 5.928 ) / (3 x 0.426) = 1.23
3S X
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 53

Reference: Example 6.2

Plot of Sample Averages

Upper contolr limit
6.5

6.3

6.1 Average

5.9

5.7

5.5 Lower control limit

5.3

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Reference: Example 6.2
Plot of Sample Range

Upper contolr limit

2.0

1.5

Average
1.0

0.5

0.0 Lower control limit

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Control Limits:
UCLR = D4 x (avg. of Range) = 2.282 x 0.991 = 2.26 (D4 = 2.282 )
LCLR = 0

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 54

Exercise 6.3 Data Type: Door Closing Effort

Sample X1 X2 X3 X4 X5 Average Range Std.Dev.

1 38.59 44.24 47.19 47.44 48.16 45.12 9.57 3.95
2 42.11 44.38 48.70 48.24 43.56
3 44.31 43.34 42.44 47.91 45.61 44.72 5.47 2.14
4 45.54 46.11 40.31 47.34 42.58
5 39.41 42.84 50.68 43.01 46.44 44.48 11.27 4.27
6 43.55 43.63 43.01 42.19 42.15 42.91 1.48 0.71
7 43.74 45.97 42.33 46.98 46.25
8 46.47 40.17 38.59 41.32 42.40 41.79 7.88 2.97
9 44.06 46.18 43.02 48.34 49.13 46.15 6.11 2.64
10 45.36 43.91 36.00 42.47 41.47 41.84 9.36 3.58
11 46.40 52.34 42.80 42.33 48.85 46.54 10.01 4.20
12 41.07 45.97 46.51 46.58 47.08 45.44 6.01 2.48
13 48.56 40.46 43.99 48.16 42.73 44.78 8.10 3.51
14 46.87 44.02 48.31 43.88 48.34 46.28 4.46 2.21
15 42.26 47.08 49.13 42.26 52.34
16 51.37 48.74 45.32 46.47 44.20 47.22 7.17 2.86
17 41.36 46.36 45.93 48.60 43.59
18 39.34 44.92 39.34 43.70 44.13 42.29 5.58 2.72
19 43.16 51.40 35.89 46.47 45.93 44.57 15.51 5.69
20 45.46 36.86 46.87 42.84 52.34

Grand Avg.= Avg.Range = Avg.Std.Dev.=

Reference:

Plot of Sample Averages

Upper contolr limit

Average

Lower control limit

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 55

Exercise 6.4 Data Type: Pressure Plate Height

Sample X1 X2 X3 X4 X5 Average Range Std.Dev.

1 5.74 5.80 5.72 5.78 5.63 5.73 0.17 0.07

2 6.17 6.29 5.69 6.14 6.14 6.09 0.60 0.23
3 7.09 6.63 5.65 6.39 6.35
4 5.80 6.15 5.96 5.97 5.48 5.87 0.67 0.25
5 6.85 5.55 6.16 6.25 5.43 6.05 1.42 0.58
6 6.19 6.29 6.37 5.84 9.60
7 5.97 6.13 5.25 5.80 5.45 5.72 0.88 0.36
8 5.51 5.71 7.09 6.19 5.70 6.04 1.58 0.64
9 5.81 5.92 5.26 5.89 5.85 5.75 0.66 0.27
10 6.72 5.70 5.54 5.99 6.50
11 6.02 5.87 5.67 5.27 6.17 5.80 0.90 0.35
12 5.80 6.08 6.51 5.83 6.17 6.08 0.71 0.29
13 5.84 5.25 5.43 5.51 5.50 5.51 0.59 0.21
14 6.33 5.68 5.49 5.14 6.27 5.78 1.19 0.51
15 6.46 6.02 6.37 7.08 6.04
16 5.27 6.14 5.74 5.71 6.25 5.82 0.98 0.39
17 5.78 5.84 5.61 6.50 5.50 5.85 1.00 0.39
18 5.78 6.20 6.09 6.27 5.46 5.96 0.81 0.34
19 5.73 4.80 6.37 6.61 5.75
20 5.61 6.74 5.57 5.81 5.80 5.91 1.17 0.48

Grand Avg.= Avg.Range = Avg.Std.Dev.=

Reference:

Plot of Sample Averages

Upper contolr limit

Average

Lower control limit

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 56

Exercise 6.5 Data Type: Stopping Distance

Sample X1 X2 X3 X4 X5 Average Range Std.Dev.

1 151 149 130 160 134 144.80 30 12.48
2 167 131 146 126 115 137.00 52 20.14
3 150 139 149 146 145 145.80 11 4.32
4 153 112 147 123 126 132.20 41 17.20
5 175 151 167 125 112
6 128 123 123 119 185
7 139 149 116 123 139 133.20 33 13.39
8 132 126 148 125 163
9 131 146 146 141 128 138.40 18 8.44
10 139 141 139 112 148 135.80 36 13.81
11 112 120 126 130 140
12 149 112 151 124 112 129.60 39 19.27
13 146 105 119 141 125 127.20 41 16.65
14 148 123 123 134 139 133.40 25 10.74
15 141 117 156 130 128 134.40 39 14.78
16 106 151 133 128 154 134.40 48 19.42
17 124 138 109 147 132 130.00 38 14.44
18 137 129 120 143 123 130.40 23 9.58
19 123 126 129 143 118 127.80 25 9.42
20 130 147 176 145 127

Grand Avg.= Avg.Range = Avg.Std.Dev.=

Reference:

Plot of Sample Averages

Upper contolr limit

Average

Lower control limit

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 57

Module 7
Interpretation of Control Charts

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 58

UCL

__ __
X or R

LCL
Time

How to read the control chart?

The purpose of examining a control chart is to detect if there is an out of control situation during
the time the data was collected.

If a process is in control, then, there will be no change in mean or range present over time. This
means that the variations in the values of individual sample means and the ranges are due only to
common causes (random behavior). If there is variation due to special causes (non-random
behavior), the process will be considered out of control.

How do we detect that there is variation due to special causes?

When the process output is affected by special causes of variation, it often produces distinct
control chart (pattern). Detecting one or more such distinct behavior derives conclusions about a
particular data plot.

Mathematics behind detection of unlikely (non-random) behavior.

Considering the normal distribution, the area under the +/- 3xStandard Deviation is 99.73%. This
means that the probability of a data point falling outside the upper or lower control limit is (100
- 99.73) = .27 %. or 2.7 parts per thousand (1.3 per 1000 on one side ). This is a good number to
remember for comparing with other probabilities.

Probability that a data point falls on one side of the mean is 0.5
Probability that two points in a row will fall on the same side of the mean is (.05)(.05) = 0.25

Probability that 9 points in a row will fall on one side of the distribution is (0.5)9 = .00195 = 2
/1000

Probability that head will show 10 times in a row is (0.5)10 = 0.097 = 0 .1% = 1/1000

Probability that you will get a queen is 4/52

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 59

Probability that you will get two queens in a row is 4/52 x 4/52 = (16/2704) =0.0059 = 6/1000

Probability that a data point will be outside 2σ limit is (100 - 95.45)/2 = 2.275% = .02275
Probability that two points in a row will be beyond the 2σ limit is (.02275 x .02275) = .0005
=.05 %

Likewise, probability that 4 points in a row will be outside the 1σ limit is (.159)4 = .000064 =
.06% =.6/1000

since, probability of one point outside is (100 - 68.27)/2 = 15.86%= .159

Who makes the rules for out of control detection?

It varies from author to author. You can create your own rule based on the probability of
occurrence.

General approach: If some situation has less that 1 in a thousand chance (1.3 per 1000 beyond
either control limits), it is a signal for unusual (non-random) circumstance.

General guidelines for detection of probable out-of-control situation. Each of the following
situations will signal the presence of special cause of variations.

Conditions Probability
of occurrence
1. A point is beyond the upper or lower control limits 0.135 %

2. Two out of three consecutive points are beyond the two standard deviation away 0.153 %
from the centerline.

3. Four out of five consecutive points are beyond one standard deviation away from 0.277 %
the centerline.

4. Six consecutive points are in an upward trend. 0.139 %

5. Eight consecutive points are above the centerline. 0.391 %

6. Eight out of ten consecutive points, taken from left to right, shows .069%
upward/downward trend.

7. Second of two consecutive point is at least four standard deviation above/below the 0.234
first one.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 60

1 2
UCL
3
7
4
__ __
X or R

5
LCL 6
Time

The number next to a point indicates the condition of out-of control

How to analyze data plots on the range charts

i) Points beyond the control limits

The occurrence of one or more points beyond either control limits primary evidence of non-
control at that point. If the process is in a state of control and only variation due to common
causes is present, any points beyond control limits would be rare.

A point above upper control limit for ranges is generally due to one of the following:
• The control limits calculated erroneously
• The piece-to-piece variability or the spread of the distribution has worsened, either at that
point or as part of the trend.
• The measurement system has changed.
• The measurement system lacks appropriate discrimination.

A point below the lower control limit (for n > 6) is generally due to one of the following:
• The control limit calculated is wrong.
• The spread of the distribution has decreased.
• The measurement system has changed.

Patterns or Trends Within the Control Limits:

The presence of unusual pattern or trends even when all ranges are within control limits can be
evidence of non-control. An unfavorable trend should be corrected. A favorable trend + should
be studied for possible permanent improvement of the process.

ii) Runs - following are signs of process shift:

• 7 points in a row on one side of the average
• 7 points in a row that are consistently increasing or decreasing

Possible causes:
-- change in measurement system

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 61

-- greater range indicates trouble and should be investigated

-- reduced range should be studied for permanent process change.

iii) Obvious nonrandom pattern -- presence of distinctive pattern offers clues to special
causes.

Distance from R - for a process in a state of control, 2/3 of the data should be within 1/3 of
the control limits and 1/3 of the data should be within the outer two thirds of the region.

If substantially more than 2/3 of the points lie close to the process average (for 25 subgroups if
90% are in the middle third of the control limits), look for the following causes.

• Control limits miscalculated

• Data have been edited
• Subgroups have been stratified and have different averages.

If significantly fewer than 2/3 of the data points lie close to the process average (40% or fewer),
suspect one of the following.

• Control limits have been miscalculated.

• The sampling method used caused subgroups to contain measurements from two or more
very different process streams. If several process streams are present they should be
identified and tracked separately.

(Course materials to be added later)

Module 8 – Module 16 Excluded from this document

Module 17

Review Topics, Exercise, and Q&A

Key Topics of Discussions (Things you should understand from this overview sessions)

(1) Distribution Functions

Normal, Binomial, Exponential, logarithmic

- Preparation
- Areas of application
- Characteristics & special properties

(2) Capability Statistics

How to compare two populations
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 62

Capability indices (calculations and implications)

(3) Statistical Process Control (SPC)

Purpose: Identify common and special causes

Technique: Create standard and compare performance with it.

- Use key properties of ND (Avg. & Std. Dev.)

- Calculate Std. Dev. Using arithmetic
- Create standard AVG. chart for stable performance
- Collect data and plot them
- Use std. Guidelines for causes of variation

Review Problems and Exercises

1. What is the common way to measure the central tendency of data?

[ ] Mean [ ] Median [ ] Mode

2. What is the most common characteristic used to represent dispersion of data?

[ ] Range [ ] Standard Deviation [ ] Variance

3. Calculate the average (a) and St. Dev. (s) of the following data and plot the population
distribution.

7 5 5 4 6 and 3

(Ans: a = 5 s = 1.414 )

4. Calculate the average (a) and St. Dev. (s) of the following data and plot the population
distribution.
9.5 8.3 8.6 9.2 9.6 10.0 8.8 and 9.3

(Ans. a = 9.162 s = 0.562)

5. Plot the normal distribution for a distribution with a = 12 and s = 2. Show the distribution
when the standard deviation, s = 1. How high would the peak of the second distribution be
compared to the first one? Calculate maximum height of distributions (use expression for
normal distribution) and label x and y-axes before plotting graph.

(Ans. Approximately twice as much)

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 63

12
6. Determine approximate value of the standard deviation of the distribution shown below.

21 24 27 30 33 36 39

7. What are the causes of variability? Which one is more common?

(Ans. Common cases, Special causes, 94%, 6%)

8. When is a process considered to be in CONTROL (or STABLE)?

Plot a time history graph of the following four situations.

[ ] No variation in Mean (a) and Spread (s)

[ ] No variation in Mean (a)
[ ] No in Spread (s)
[ ] There are variations in Mean (a) and Spread (s)

9. Plot the graph for the following scenarios (use separate pages to draw 2-d and 3-D graph with
time axis).

A process that is:

(a) STABLE and within SPECS.
(b) STABLE but out-of-SPECS.
(c) STABLE, out-of-SPECS, but CAPABLE.
(d) STABLE, out-of-SPECS, and INCAPABLE.
(e) UNSTABLE but in-SPECS.
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 64

(f) UNSTABLE and out-of-SPECS.

(g) UNSTABLE but improving mean
(h) STABLE but improving spread

10. Hotel operator Mr. Marriott routinely fills most of his 800-room hotel in a major
metropolitan city. Analysis of the occupancy data (number of rooms sold each night) for past
several years show average occupancy of 700 rooms and a standard deviation of 12 rooms.
For last few weeks, the hotel’s occupancy jumped to 725 (average occupancy). Should Mr.
Marriott initiate a hotel expansion project?

Process Capability Indices (Cp & Cpk)

11. A newly installed electricity generating plant is found to operate with the following data:
Average voltage =112 volts and Standard Deviation = 2.93 volts. The target voltage is 110
and the specification limits are USL = 120 and LSL = 110 volts. Calculate the generator
process potential (Cp) and the process capability (Cpk).

(Ans. Cp = 1.137 Cpk = 0.909)

12. Plot a typical data/distribution and label USL, LSL, Target, and Mean of the data. Also, write
the expressions for Cp and Cpk. Answer following questions.

(a) When is Cpk same as Cp?

(b) How can Cp value be increased?
- Increase (USL-LSL)
- Decrease standard deviation (s)
- Increase (USL_LSL) and decrease s

(c) What is the desirable location of the mean for better Cpk?
(d) If you want over 99% good production part, what should the Cpk of the process be?
(e) How are Cp and Cpk related? Can Cpk ever exceed Cp?
(f) How can Cpk value be increased?
- Move mean closer to target
- Reduce standard deviation
- Move mean closer to target and reduce standard deviation
- All the above plus, wide USL/LSL further

13. The foreman of a machine shop was to select the best process among the two competing
production line producing resistors for TV-appliance. The following specifications and
production data were available:

Target value = 75 Ohms, LSL = 55 USL = 95

Measured data from sample production output:

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 65

Line 1: 66 76 83 80 73 75 84 and 69 ( n = 8)

Line 2: 74 79 67 68 75 74 76 67 and 71 (n = 9)

(Ans: a = 75.75 s = 6.408 a = 72.333 s = 4.301, cp=1.040 Cpk =1.00 Cp =1.55 Cpk=
1.343 )

14. All things remaining the same, if the standard deviation of the process parameter is reduced
to its 50% level, what will happen to the Cpk index?

15. What happens to the Cpk index when the specification limits are arbitrarily narrowed to half
of the current range?

Construction of Control Charts (SPC)

QA:
Q1. What are control charts?
Ans: Control charts are powerful tools that can be used to monitor the performance when
appropriate measurements from a manufacturing or service delivery process are available.

Q2. What are typical measurement data for which control charts (X-bar and R charts are most
common) are appropriate?
Ans: The data must be of variable type. Typical measurements are: Length (in centimeters,
inches, etc.), Width, Diameter, Time, Strength, etc.

Q3. What are X-bar and R charts any way?

Ans: X-bar and R charts are nothing more than plots of process X-bar and R over time, with two
additional lines (UCL & LCL) drawn on the plots. The two additional lines are called the control
Limits.

Q4. What are benefits of the control charts (Say for X-bar and R charts)?
Ans: A variable data control chart is used to determine a process performance characteristic in
terms of both its location (mean of the characteristic) and its spread (variability of the
characteristic). For this reason, control charts for variable data are used in pairs – one chart (X-
bar chart) to assess centrality and the other to evaluate process spread (R chart).

Q5. How do control charts indicate if the process is out of control?

Ans: Control chart represents the normal or standard process performance. Obviously, control
charts are constructed by collecting data from the process when it is STABLE. Current and future
performance data then are compared against the expected normal behavior. Observation about
abnormal behavior then can be made by comparing the performance with statistically valid
characteristics of normal distribution (99.7% within USL & LSL).

Q6. What is a common data sampling strategy?

Ans: For both construction of control chart and monitoring of performance, a common practice is
to collect data 25 samples of data with 5 data in each sample (5x25 readings). Average of each
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 66

sample is X-bar and range of a sample is R. The average of all sample averages is called X-
double-bar and the average of sample ranges is called R-bar.

Q7. Is standard deviation for 125 samples needed for the control chart?
Ans: Because it is difficult to directly calculate standard deviation of the total data (125),
approximate, but simpler method is used to calculate standard deviation. For control charts upper
and lower control limits are usually calculated from R-bar.

Q8. What are plotted in the X-bar chart?

Ans: After the control chart is constructed (with X-double-bar, UCL, and LCL), X-bar of each
collected sample (of 5 data) are plotted in the X-bar chart.

Q9. Why are the upper and lower control limits (LCL & UCL) of X-bar chart set at three
standard deviation of X-bars apart.
Ans: Three standard deviations from the mean are expected to contain 99.7% of the data.

Q10. What are the steps in constructing X-bar control chart?

Ans: Step1 – Collect Production data (variable data, 25 sets of 5 data in a set.), Step 2 – Record
data in a table form, Step 3 – Calculate X-bar and R for all 25 samples, Step 4 – Calculate X-
double-bar and R-bar, Step 5 – Calculate UPPER and LOWER CONTROL LIMITS, Step 6 –
Scale the vertical axis of the x-bar chart, Step 7 – Plot the values of the X-bar in the control chart,
Step 8 – Draw the center line for X-double-bar, and Step 9 – Draw the control limits.

Q11. How are control limits calculated?

Ans: Control limits are calculated as follows.

UCL (for X-bar) = X-double-bar + 3 x Std. Dev. Of X-bars

UCL (for X-bar) = X-double-bar - 3 x Std. Dev. Of X-bars

Where Std. Dev. Of X-bars = (Std. Dev. Of X )/Sq.Root of n (n = 5)

Also Std. Dev. Of X = R-bar/d2 (d2 = 2.326 for n =5)

Or Std. Dev. Of X-bars = Constant x R-bar

UCL (for X-bar) = X-double-bar + A2x R-bar (A2 = 0.577 for n =5)
UCL (for X-bar) = X-double-bar - A2x R-bar
(See class notes or seminar materials for formula notations)

Q12. What does it mean when we say a process is out of control?

Ans: There are a number of ways a process can be out-of control. Out-of-control means that over
time there are changes in either the center of the process or the spread of the process.

Q13. What is the nature of common changes?

Ans:
Gradual change in the center or the spread (worker fatigue, change of working condition, ),
Abrupt and persistent …(change of operator, reaction to new setup, ),
R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509
 Page 67

Abrupt but temporary change…,

Periodic movement in …(weekly environmental changes, operator rotation, ),
Erratic or systematic changes in the center or the spread (due to special & specific causes).

Q14. What are some of the common rules for detecting out-of-control patterns?
Ans:
a. Single point outside the control limits
b. Three consecutive points between 2 & 3 Std. Dev.
c. Seven points in a row on one side of the X-double-bar line (Run)
d. Six points in a row either increasing or decreasing (Trend)

Ref: Normal Distribution

+/- s = 68.27%, +/- 2s = 95.45%, +/- 3s = 99.73%, +/- 4s = 99.9937%
1 ( x−µ ) 2

f ( x) = e 2σ 2 ....− ∝≤ x ≤ + ∝
σ 2∏

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509

 Page 68

Program Evaluations

Program Title_________________________________________________________________________________
Program completion date

Instructor Ranjit K. Roy & …. Training Location/host.____________________________________

We appreciate your comments and suggestions. Please take a moment to let us know how we can improve and serve you
better. (Please use the following numbers for evaluation purposes).

[6] Excellent [5] Very Good [4] Satisfactory [3] Poor [2] Unsatisfactory [1] Needs Improvement

1. Overall Reaction to Program: 6 5 4 3 2 1 (Circle one number)

Comments:_________________________________________________________________________________________

__________________________________________________________________________________________________

2. Reaction to Instructor: 6 5 4 3 2 1 (Circle one number)

Comments:_________________________________________________________________________________________

__________________________________________________________________________________________________

__________________________________________________________________________________________________

3. What should be added and/or deleted to improve this program? (Please be specific)
__________________________________________________________________________________________________

4. What portions of this program do you feel will be most helpful at your work? How?
__________________________________________________________________________________________________

__________________________________________________________________________________________________

5. Do you feel the training provided you enough understanding of the technique for you to be able to start applying it to your
own projects if opportunities were available?
[ ] Yes [ ] No [ ] May be

6. Would you consider/recommend us for training at your facility? [ ] yes [ ] No [ ] May be

Who should we contact for training at your facility _______________________ Ph: _____________________

7. Would you recommend that others attend this seminar? [ ]YES [ ]NO

8. (Optional)
Your Name________________________ Your Ph#/ & E-mail:_________________________________________
THANK YOU for taking the time to complete this evaluation.

R. Roy/Nutek, Inc. Statistical Process Control – Management Overview Version 0509