Statistical Process Control

V.V.Iyer

Aathreya Consultants - VVI

1.0 Introduction.

Aathreya Consultants - VVI

What
1.0 is SPC?
INTRODUCTION

SPC stands for Statistical Process Control.

Aathreya Consultants - VVI

What is a Process?

Inputs PROCESS Outputs

A process can be described as a transformation of set of inputs into

desired outputs.

What is Control?

Comparing the actual performance with the requirement and

deciding when and what corrective action is to be taken.

What is Statistics?

Statistics deals with data collection, summarization, analysis,

conclusion and decision making.
Aathreya Consultants - VVI
Statistical data are collected from the process
and analyzed to understand the process behavior,
predict the process performance and decide the
course of action.

Aathreya Consultants - VVI

 Fundamental Principles of SPC

1. No two things are exactly alike.

2. Variation in any process or product can be measured and

quantified.

3. Things vary according to a definite pattern.

4. It is possible to mathematically model the variation pattern in

the output of any process.

Aathreya Consultants - VVI

 The main focus in SPC is to quantify the variation
understand its sources and find ways and means of
reducing the variation.

Measure the output of the process

Analyze the output measurements and summarize

From the sample measurements, generalize to overall

output of the process (model building)

Verify whether the outcome of such analysis is

satisfactory and adequate

ok? Yes Monitor the output

continuously
No
Investigations andAathreya
improveConsultantsthe
- VVI process
Quality characteristic is the output characteristic
which we want to measure and improve.

Measuring means expressing the quality characteristic in

numbers (numerical form).

Aathreya Consultants - VVI

What is Data?

Data is the numerical expression of a phenomenon or process

output.

Sources of Data

 Past records
 Live data.

Aathreya Consultants - VVI

 Types of Data

Data on Quality Characteristic

Measurable Not Measured / Attribute

Length Ok / Not ok
Weight Good / Bad
OD Defective / Non-defective
Hardness Present / Not present
Tensile Strength
Aathreya Consultants - VVI
2.0 Summarising Data.

Aathreya Consultants - VVI

2.0 DATA SUMMARIZATION
To evaluate a process, process data must be collected and
summarized.
Summary of Data

Graphical Summary Numerical Summary

- through Graphs & Charts - through Process Measures

For variable
* Bar Chart * Measures of central tendency
* Line Chart * Measures of dispersion
* Pie Chart * Measures of skewness
* Histogram * Measures of kurtosis

For attribute
- Number of defective units
- Number of defects
Aathreya Consultants - VVI
2.2 Numerical Summary – Process Measures

Summarisation means condensing a set of process data into

a few representative measures. It

 facilitates easy understanding

 enables easy comparison

 Makes it easier to generalize to overall population

 helps to make quick decision

Aathreya Consultants - VVI

Process Measures for Variable Data

Measure of Central Tendency

In any process data, it can be observed that most of the

data tend to cluster around a central value. This
phenomena is known as central tendency and the value
around which data is clustered is called the measure of
central tendency.
Central tendency is measured through averages.

Commonly Used Averages

 Arithmetic Mean (X-bar)

 Median (M)
Aathreya Consultants - VVI
Arithmetic Mean (X-bar)

It is defined as the ratio of the total of all the observations

to the total number of observations.

It is computed as

X-bar = 1/n (X1 + X2 + … + Xn)

X-bar = 1/n ni=1Xi

where X1X2 , … Xn, are individual observations. n:

number of observations.
Aathreya Consultants - VVI
Median (M)

It is defined as the middle most value in the data set.

To compute median, the data are to be arranged in the

increasing order and the middle value is to be picked up.

Aathreya Consultants - VVI

Example

Compute the Arithmetic Mean (X-bar) and the median for

the following data set.
7, 10, 8, 6, 4

Arithmetic Mean X-bar = 1/5 {7 + 10 + 8 + 6 + 4}

= 1/5 x 35
=7

Median M = 4,6,7,8,10 arranged in increasing order

= Middle Value = 7.
Aathreya Consultants - VVI
Measures of Disperson

It is a measure of spread or variability in the data.

Commonly Used Measures of Disperson

 Range (R)

 Standard Deviation (s)

Aathreya Consultants - VVI

Range (R)

It is defined as the difference between the largest and

smallest value in the data set.
It is computed as
R = Xmax – Xmin

Standard Deviation (s)

It is defined as the root mean squared deviation from the

average X-bar.
It is computed as
s = +  { [1/(n-1)] (Xi – X-bar)2
where Xi – individual value
X-bar – Arithmetic Mean
n – number of -observations.
Aathreya Consultants VVI
 Standard deviation is a better measure than Range.

eg.Compute the range (R) and standard deviation (s) for the
following data set.
7, 10, 8, 6, 4

Range (R) = Xmax – Xmin

= 10 – 4
=6

Standard Deviation (s)

= +  {(1/4) [(7-7)2+(10-7)2+(8-7)2+(6-7)2=(4-7)2]}
= +  {(1/4) [0 + 9 + 1 + 1 + 9]}
= +  {(1/4)(20)}
= + 5 = 2.236. Aathreya Consultants - VVI
Process Measures for Attribute Data

For defective items, the process performance is quantified

by average proportion defective p-bar.
It is computed as

p-bar = [d1 + d2 + … + dk] / [n1 + n2 + … + nk]

= [ki=1di] / [ki=1ni]
where, k is the number of subgroups

di is the number of defective items in the ith subgroup

ni is the size of the ith subgroup.

Aathreya Consultants - VVI
Process Measures for Attribute Data (Contd…)

For defects data, the process performance is quantified by

average number of defects per unit c-bar.

It is computed as

c-bar = [ki=1ci] / k

where, k is the number of units sampled and

ci is the number of defects observed in the ith unit

Aathreya Consultants - VVI

3.0 Normal Variation Pattern
(Normal Distribution)

Aathreya Consultants - VVI

 Normal Variation Pattern
(Normal Distribution)

It is the most popularly used statistical distribution to model

the variation pattern in a measured quality characteristic.

Aathreya Consultants - VVI

Histogram of sample dimensions

Aathreya Consultants - VVI

Histogram and distribution curve of sample dimensions

Aathreya Consultants - VVI

Distribution curve of sample dimensions.

Aathreya Consultants - VVI

 Properties of Normal Distribution

 Model for variable data

 The frequency curve is bell shaped

 The distribution is symmetrical around average

 The distribution is completely known when its average (x-

bar) and standard deviation (s) are known.

Aathreya Consultants - VVI

 Normal Distribution

–
X

- 3 - 2 - 1 68.26 % 1 2 3
95.45 %

99.73 %

Aathreya Consultants - VVI

 Standard Normal Curve
Instead of dealing with various normal distributions with
different averages and standard deviations, they can all be
reduced to a common form known as the standard normal
distribution.
Y

-3 -2 - 1Aathreya Consultants
0 - VVI1 2 3
 Standard Normal Curve

Tables have been constructed to find out the area under

the curve between any two given points z1 and z2 in the
standard normal curve.

This is called the standard normal table.

Any probability statement regarding any normal distribution

can be evaluated using these tables.

In particular, the expected non-conformance (PPM level) in

a process can be estimated using this table.
Aathreya Consultants - VVI
4.0 SPC – Basic Concepts

Aathreya Consultants - VVI

 Functions of a Process Control System are

 To signal the presence of assignable causes of variation

 To give evidence if a process is operating in a state of statistical

control

 If the process is in a state of statistical control, meaningful

assessment of its capability to meet the specifications can be
made.

Aathreya Consultants - VVI

Voices of Two Processes, One Unstable and One Stable

Aathreya Consultants - VVI

A Stable Process Might Not be Normally Distributed

Aathreya Consultants - VVI

 Benefits of a Control Chart

 Gives signal before the process starts deteriorating

 Aids the process to perform consistently and
predictably
 Gives a good indication of whether the problems are
due to operational faults or system faults
 Provides a common language for communicating
the performance of any process
 Provides a record of process performance
 Simple to use and can be maintained by the
operator himself.

Aathreya Consultants - VVI

 Types of Data

Variable data
Data that takes any value between –  and + 

Example: Diameter of a shaft 20.8976, 20.89785, etc.

Thickness of a plate

Attribute data
Data that takes only discrete values

Go – No go; Conforming – Non Conforming; Accept – Reject

No. of defective components
No. of defects in a set of components.
Aathreya Consultants - VVI
 Essential features of a control chart

 Central Line (CL)

 Upper Control Limit (UCL)
 Lower Control Limit (LCL)

 Control Limits are calculated from the process output

data
 Control Limits are not Specification Limits
 Specification Limits are provided by the Engineering
Department based on the functional requirement
 Control Limits are decided based on the process
behavior.
Aathreya Consultants - VVI
Essential features of a control chart (Contd…)

Upper Control Limit

Variable Values

Central Line
Lower Control Limit

Time

Aathreya Consultants - VVI

 Control Limits are calculated based on the assumption
of normal behavior of the process
 Control Limits are placed at 3 times the Standard
Deviation of the characteristic plotted from the process
center
 Values beyond Control Limits signify the presence of
abnormality in the process factors, i.e., the presence of
assignable causes.
Characteristic
Values

Time Aathreya Consultants - VVI

 Applications

 Most commonly used Control Chart is Average & Range

Chart (X-R Chart), with typical subgroup size 4 or 5 or 6

 When subgroup sizes are to be increased beyond 9, the

control chart to be used is Average and Standard
Deviation Chart (X-S Chart)

 Median Charts are used if the calculation of Average is

difficult for the operator

 Moving Range Charts are used when the rate of

production is low and only single observations are
possible
Aathreya Consultants - VVI
 Applications (Cond…)

 Pre-Control Charts are essentially plotting individual

observations, with the tolerance zone divided into zones
of action to be used only if the process capability is
above 2

 When the process variability is very low compared to the

characteristic tolerance, the control limits can be
relaxed as in Modified Control Charts

 Other Charts are Control Charts for special applications.

Aathreya Consultants - VVI

 Types of Control Charts for Attribute Data

Description Type Sample Size

Control Chart for proportion p Chart May change

non conforming units
Control Chart for no. of non np Chart Must be constant
conforming units in a sample
Control Chart for no. of non c Chart Must be constant
conformities in a sample
Control Chart for no. of non u Chart May Change
conformities per unit

Aathreya Consultants - VVI

Selecting the Most Appropriate Control Chart

Aathreya Consultants - VVI

5.0 Control Chart for Variables

Aathreya Consultants - VVI

 Steps to construct Average – Range Chart

 Generate a sample size of 4 or 5 at periodic interval

(called sub group)
 Compute the sample Average and the Range
 Repeat the above for at least 25 times, preferably over
all the three shifts
 Compute the average of the ranges (R-bar)
 The Control Limits for R chart are
 Center Line (CL) : (R-bar)
 Upper Control Limit (UCL) : D4 * (R-bar)
 Lower Control Limit (LCL) : D3 * (R-bar)

Aathreya Consultants - VVI

 Steps to construct Average – Range Chart

 Homogenize R Chart
 Plot all the Range values in the R Chart
 Remove those values that do not lie within the UCL and LCL
 Compute R-bar with the modified data
 Compute new UCL and LCL with the new R-bar
 Repeat above steps till all the Range values are within the
new UCL and LCL.
 Compute overall Average (X-double bar)

Aathreya Consultants - VVI

 Steps to construct Average – Range Chart

 The Control Limits for X-bar chart are

 Center Line (CL):(X-double bar)
 Upper Control Limit (UCL):(X-double bar)+A2*(R-bar)
 Lower Control Limit (LCL):(X-double bar)-A2*(R-bar)
 Homogenize X-bar Chart
 Plot further points for future control, using the established Control Limits in
both the Charts
 If the process is not centered during the initial construction of the chart,
 the UCL and LCL of the X-bar Chart can be calculated for future control
as,
 {Nominal +/- A2 * (R-bar)} respectively,
 with CL as the nominal
 after centering the process.

Aathreya Consultants - VVI

 X-bar & S Chart

Applicable if the sub group sizes are large (about 10)

Construction procedure is similar to X-bar & R Chart
For each sub group, the sample Standard Deviation (S) are to be calculated
The average of all the sample Standard Deviations is S-bar
The Control Limits are calculated as

Chart CL UCL LCL

X-bar Chart X-double bar CL + A3 * S-bar CL - A3 * S-bar

S Chart S-bar B4 * S-bar B3 * S-bar

Aathreya Consultants - VVI

Control Chart of a Stable Process. The Distribution has the Same Average,
Variation and Shape Over Time

Aathreya Consultants - VVI

Lack of Statistical Control is indicated by a Control Chart by

 A point falling outside the control limits

 An increasing trend

 or A decreasing trend

 A run of 7 or 8 consecutive points on only one side of the Central Line

 A cyclic pattern
Aathreya Consultants - VVI
Control Chart of a Process with Unstable Variation. Because the
s Chart is out of Control, the Control State of the X Chart Cannot
be Interpreted

Aathreya Consultants - VVI

6.0 Process Capability Analysis

Aathreya Consultants - VVI

 Process Capability

 Process capability of a process is the extent of variation in the

Quality characteristic of the process output, when the process is
operating in a state of Statistical control under a given set of
conditions

 Concerns with the ability of the process to produce output meeting

specifications consistently

 Takes into account variation caused by all the possible sources of

variation

 Data for Process Capability Analysis should be collected over a

period of time

Aathreya Consultants - VVI

Aathreya Consultants - VVI
 Process Capability

 Process factors leading to variation are

 Raw Material
 Machine/Equipment
 Accessories
 Setting
 Interruptions
 Operator
 Measuring system
 Environment

Aathreya Consultants - VVI

 Uses of Process Capability Study

 Provide Information to facilitate design of product

characteristics/tolerances
 Assist process planners to select or modify a process
 Assist in establishing the interval between sampling for Process
Control
 Specify the performance requirement of a new process
 Compare the capability of various processes
 Selection of operators
 Assigning work to machines
 Selecting between competing vendors
 Determining the economic nominal for an operation
 Reducing the variability in a process.

Aathreya Consultants - VVI

 The variation can be traced to two basic sources.
Variation

Common Cause Assignable Cause

Inherent to process Not inherent

Due to many random Due to few specific
unidentifiable sources identifiable causes
Magnitude is small Magnitude is large
Follows predictable pattern
Doesn’t follow any pattern /
Can’t be reduced without
Sporadic.Can be eliminatedy
process redesign or high cost

The Process Capability is then described as the Total Common Cause

Variability Present in a Stable Process.

Aathreya Consultants - VVI

Process Capability for Normal Distribution

Process capability is the minimum variation that has to be tolerated after all the
special causes of variations have been eliminated.

If a process for which the quality characteristic has a normal distribution the
process capability is measured as 6 x s.d (6s).

Aathreya Consultants - VVI

Process Capability Analysis
There are two distinct approaches used to analyze Process capability.

Process Capability

Short run capability Long run capability

Concerned with variability Concerned with variability

over a short period of time over a larger period of time

Prerequisites for Process Capability study:

• Stability of the process

• Acceptable measurement system

Aathreya Consultants - VVI
 Visualizing Short-Term and Long-Term Variation –
"Individual World"

Long-Term

• • Long-Term

• •• • •• • • Short-Term
Short-Term
Measurement

• • ••

Measurement
• •• • • •• •• • • • • • ••
• • • •• • • • •• •
• • • •• •
Time
•
Time

Long-Term

•• • Short-Term Short-Term Long-Term

Measurement

Measurement

• • •• ••• • •• • •• • •• • • •• • •
• ••• •
• •• • • ••• • • ••• • • • •
•
Time Time
Aathreya Consultants - VVI
 Short-Term vs Long-Term Variation
Long-term (LT)
Short-term (ST)

•Represents the process •Represents the total

capability process variation

•Captures variation due to •Captures variation due to

common causes common and special
causes
•Measures variation within
a subgroup or between
successive values
•Measures variation in all
data
•Used for calculating control
chart limits •Should not be used for
calculating control chart
limits
Aathreya Consultants - VVI
Measuring Potential Capability with Bilateral Tolerances

CP is a measure of short-term potential capability, and PP is

a measure of long-term potential capability. Here are the
definitions of CP and PP:

Aathreya Consultants - VVI

 Process Capability Indices

 Process Capability Indices are to be calculated only for those

processes in Statistical Control
 X-bar and  can be estimated from the data collected from Control
Charts
 Fairly large number of readings spanning a wide time period should
be taken for consideration
 Process Average is the overall average of all the observations
considered, i.e., X-double bar
 Process Standard Deviation  is estimated as,
 = (R-bar) / d2 , in case of X-bar & R Chart

Aathreya Consultants - VVI

 Calculating Short-Term and Long-Term Standard Deviations

Short-term (ST) Long-term (LT)

ST = R / d2 
n
Xi – X 2
LT = i=1
n–1
ST = s / c4

ST = MR / d2

If Stable ,  LT   ST
If Unstable,  LT   ST
Aathreya Consultants - VVI
 Stability Index

LT
Stability Index = 
ST
For a stable process, you would expect index values near 1.
For an unstable process, you would expect index values greater than 1.

Rule of Thumb
< 1.33 – Good Process Stability.
1.33 to 1.67 – Marginal Process Stability.
> 1.67 – Major Process Stability Issues.

Note: For use when n>75. If n<75, consider using <1.5, 1.5-2.0, and
>2.0.
Do not use when n<30
Aathreya Consultants - VVI
Examples of Eight Processes with Potential
Capability Metrics

Aathreya Consultants - VVI

Target, T = 18.0

Upper Specification Limit, USL = 36.0

Lower Specification Limit, LSL = 0.0

Aathreya Consultants - VVI
 Average, Process Spread
X-bar 6s
Process A 18.0 (28 – 8) = 20.0

Process B 26.0 (36 – 16) = 20.0

Process C 36.0 (46 – 26) = 20.0

Aathreya Consultants - VVI

 Tol. = 6s Cp =
(USL - LSL) Tol. / 6s
Process A 36 20 36 / 20 = 1.8

Process B 36 20 36 / 20 = 1.8

Process C 36 20 36 / 20 = 1.8

Aathreya Consultants - VVI

Weakness of Cp

CP measures only the potential of a process to produce

acceptable product.

CP gives no indication of actual process performance .

Aathreya Consultants - VVI

Cpk Index is defined as

CPK = min {CPU , CPL}

The CPK is estimated from the sample data as


USL  X X  LSL
C PK
 Min{ , }
3s 3s
A process is considered as capable if CPK > 1.0.

Aathreya Consultants - VVI

 Cp Cpu Cpl Cpk

Process A 1.8 = (36-18)/10 = (18-0)/10 1.8

= 1.8 = 1.8

Process B 1.8 = (36-26)/10 = (26-0)/10 1.0

= 1.0 = 2.6

Process C 1.8 = (36-36)/10 = (36-0)/10 0.0

= 0.0 = 3.6

Aathreya Consultants - VVI

 Cp Cpk NC%
Process A 1.8 1.8 0.0

Process B 1.8 1.0 0.135

Process C 1.8 0.0 50.0

Aathreya Consultants - VVI
Any capability index alone cannot fully describe the process
behavior. The indices CP and CPK must be considered
together for characterizing a process.

CP
CPK HIGH LOW
HIGH Aim target Not possible

LOW Move process mean towards Improve process

target by reducing
variability

Aathreya Consultants - VVI

Eight Example Processes with Potential and Actual Capability
Metrics

Aathreya Consultants - VVI

Acceptable values for Process Capability Indices

Critical Non Critical

Characteristic Characteristic
New Process Cpk > 2 Cpk > 1.66

Old Process Cpk > 1.66 Cpk > 1.33

Aathreya Consultants - VVI

MINITAB Capability Report

Aathreya Consultants - VVI

7.0 Control Chart for Attributes

Aathreya Consultants - VVI

 Construction Procedure

Construction procedure is similar to control chart for

variable data

 Collect data
 No. of non-conforming units in a sample
 No. of non-conformities in a sample
 Calculate control variable, as required
 Proportion of non-conforming units
 Average no. of non-conformities per unit
 Calculate Control Limits
 Homogenize data
 Calculate Control Limits for the homogenized data
 Use the Control Limits for future control.

Aathreya Consultants - VVI

 p-Chart

 Generate a sample of size n (say 25 or 30)

 Count the number of non-conforming items in each
sample (say d)
 Calculate the control variable as Proportion defective
in the sample: p = d / n
 Repeat the above steps for k samples (25 at least)
 Calculate Control Limits as
 CL = p-bar {(Total number of defectives) .
(Total number of items inspected)}
 UCL = p-bar + 3 * square root {[p-bar*(1 - p-bar)] / n}
 LCL = p-bar - 3 * square root {[p-bar*(1 - p-bar)] / n}.
Aathreya Consultants - VVI
 np-Chart

 Data collection similar to p-chart

 No need to calculate proportion defective p
 Control Variable is number of non-conforming units: d
 Control Limits are
 CL = d-bar {Mean of all d’s}
 UCL = d-bar + 3 * square root {d-bar * [1 – (d-bar/n)]}
 LCL = d-bar - 3 * square root {d-bar * [1 – (d-bar/n)]}.

Aathreya Consultants - VVI

 c-Chart

 To control the number of non-conformities in one

single inspection unit
 Ex: A unit area of sheet metal
 Count the number of non-conformities in the
inspection unit (say c)
 Control Variable is number of non-conformities in the
inspection unit: c
 Repeat the above steps for k units (25 at least)
 Control Limits are
 CL = c-bar {Mean of all c’s}
 UCL = c-bar + 3 * square root {c-bar}
 LCL = c-bar - 3 * square root {c-bar}.
Aathreya Consultants - VVI
 u-Chart

 To control the average number of non-conformities per

item in a given sample size n
 Count the number of non-conformities in the sample of
n items, say x non-conformities
 Calculate Control Variable as the average No. of non-
conformities per item: u = x/n
 Control Limits are
 CL = u-bar {Total no. of non-conformities / Total no.
of units inspected}
 UCL = u-bar + 3 * square root {(u-bar) / n}
 LCL = c-bar - 3 * square root {(u-bar) / n}

Aathreya Consultants - VVI

 u-Chart

For p-chart and u-chart

 If the sample size is changed from one sample to

another, Control Limits have to be calculated for each
sample, both during homogenizing and future control

 Alternatively, if the sample sizes are within 15% of the

average ‘n’ (n-bar) and the sample sizes ‘n’ are large,
 Control Limits with average ‘n’ (n-bar) can be used
instead of calculating Control Limits every time.

Aathreya Consultants - VVI