4/2/2010

Statistical Quality Control

Dr. Prasun Das

Indian Statistical Institute
SQC & OR Unit, Kolkata
dasprasun@rediffmail.com

April 2, 2010 1

Quality - Definition
ƒ Fitness for use (with five major dimensions – design,
conformance,, availability,
y, safetyy and field use).
) [[J.M.
Juran]
ƒ Conformance to specifications. [Philip Crosby]
ƒ Totality of features and characteristics of an entity that
bear upon its ability to satisfy stated or implied needs. [ISO
9000]
ƒ Defined by the customer; customers want products and
services that, throughout their lives, meet customers’ needs
and expectations at a cost that represents value. [Ford]

April 2, 2010 2

1
 4/2/2010

DEFINITION

Process: A combination of equipment, materials, peoples,

method and environment that act together to produce an output.
Th quality
The lit off theth process isi judged
j d d by b the
th quality
lit
(characteristics) of the output.
Control: The planned cycle of activities by means of which we
achieve stability in some specified aspect.
Process Control: The set of activities to maintain the variation in
the output characteristic of a given process within its naturally
capable limits of variation ensuring stability in the
characteristics.

2-Apr-10

DEFINITION

Statistical Control: A process is under statistical control when

the variation in output characteristic(s) is governed by
chance
h cause only
l – being
b i freed
f d from
f assignable
i bl causes.

Statistical Process Control (SPC): It is the application of

statistical methods in manufacturing and service processes of
an organization which process achieves control. Since,
statistical methods and principles are used extensively in
process control work, hence the name Statistical Process
Control (SPC).

2-Apr-10

2
 4/2/2010

Concept of Variation: NELSON’s FUNNEL EXPERIMENT

April 2, 2010 5

Natural Variation
In every process, certain amount of variability will always exist,
which is due to the natural fluctuations in many factors (like raw
materials processing parameters,
materials, parameters operating conditions and skill of
operators) affecting the process performance,. This variability itself
is a characteristic of the process and is known as variability due to
Common (Chance) Causes. Reduction of natural variability calls
Breakthrough action.

Whenever some parameter(s) in the existing cause-system change

significantly, the process does not remain ‘natural’ and it exhibits
variation larger than the limits imposed by the natural variability.
These causes are known as Special (Assignable) Causes. Reduction of
unnatural variability calls Control action.
April 2, 2010 6

3
 4/2/2010

Common Cause and Special Cause

Chance (Common) Cause

• Very large in number
• Each causes minute amount of variation
• Non economical or difficult to eliminate
• Variation follows predictable pattern

Assignable (Special) Cause

• A few in number
• Each causes large variation
• Economical and easy to eliminate
• No predictable pattern

April 2, 2010 7

Stable Process

• A process is said to be stable in case it fluctuates randomly around

a mean value with nearly same variability over a period of time.

• A stable process has a repeatable pattern of variation known as the

distribution. In a stable process, only chance causes of variation are
present, resulting in random process variation.

• Only a stable process has predictable characteristics like process

capability, average, percentage defectives etc.

• When a process is unstable, one must act on the system to bring it

back to stable state.

April 2, 2010 8

4
 4/2/2010

Natural Pattern of Variation

A stable process is governed only by common causes, the

plotted points under such situation behave in the following
ways:

¾ Most of the points are centered around the central line.

¾ A few of the points are spread out and approach the control
limits.
¾ None of the points (or at most only a very rare and
occasional point) violate the control limits.

April 2, 2010 9

Control Chart – What it is?

Control Chart is a better on-line substitute to judge whether the
process is under statistical control. It helps to take necessary
corrective action either to control or to improve
p the p
process
performance and thereby looking for a higher product (service)
quality.

April 2, 2010 10

5
 4/2/2010

Control Chart – What it is?

It is a chart on which the measured values of the process parameters /
product characteristic (temp., pressure, hardness, gcs etc.) being controlled
are pplotted in sequence.
q The chart consists of a central line (p (process
average) and two statistical limit lines called LCL and UCL which indicate
the limits of natural variation (due to chance cause) for the sample
‘statistics’ (like average, range, %defective, no. of defectives per sample, no.
of defects per item etc.) being plotted.

April 2, 2010 11

Control Limits
The control limits are usually placed at (Mean ± 3 standard error).
Standard error is a function of standard deviation of individual
observation. The standard deviation represents the variability within
a rational subgroup. Typically, for 3σ limit, if the sample statistic is
assumed to have an approximately Normal distribution, there is a
probability of only 0.0027 of a computed sample statistic falling
outside the control limits if the process is in control.

Remember: by Central Limit Theorem (CLT)

1. Sampling distribution of the sample mean is approximately normal.
2 Mean off the sampling
2. i distribution
i i i off the sample mean (X) ( ) is
i equal
to the population mean (µ).
3. Standard deviation of the sample mean is given by σ X = σ / n (i.e.,
inversely proportional to the square root of the sample size)

April 2, 2010 12

6
 4/2/2010

Theoretical Basis of Control Charts

April 2, 2010 13

Control Limits – Two Kinds of Error
The control limits are supposed to strike a balance between two kinds
of errors viz. (1) looking for trouble that does not exist and (2) failing
to look for trouble that does exit.
exit Neither of these kinds of errors
should be unduly large, yet neither should be reduced to such an
extent that it unduly increases the other.

Type I error (α): results from inferring that a process is out of control when it is
actually in control.
Type II error (β): results from inferring that a process is in control when it is
actually out of control.
control

Example 6.2

April 2, 2010 14

7
 4/2/2010

Rational Subgroup and Sampling Frequency

Rational subgroup is a sample that represents a homogeneous

group. Assignable causes, if they exist, cause variation between
groups The objective of control chart technique is to check whether
groups.
the variation between groups measured by subgroup difference is
significant over the variation within subgroups.

Sampling Frequency: Samples constituting rational subgroups are

taken at intervals. Two options for sampling scheme – large samples
((size)) with less frequent
q intervals or, small sample
p ((size)) with more
frequent intervals. The type of inspection (destructive, non-
destructive), state of the process (stable / unstable) and cost of
sampling and inspection are the major concern of sampling frequency.

April 2, 2010 15

Unnatural Pattern of Variation

An unstable process is governed both by chance (common) and

assignable (special) causes, the plotted points show the
following patterns:

¾ points violating Control limits – out of control (level, variability).

¾ run of points within control limits - sustained shift in level.
¾ trend of points - gradual change in level.
¾ Appearance of cycles – periodic influence of factor(s).
¾ Points close to UCL / LCL – two or more overlapping distributions.
¾ Points too close to central line - incorrect rational subgroup.

Read Section 6.5

April 2, 2010 16

8
 4/2/2010

Types of Control Charts

Sl. Control Nature of Process
Examples
No. Chart data performance

Dimension, Height,
Width, Resistance,
Measurable Average and Strength, Tapping
1. X–R
(Variable) variability temp., Waiting time,
Service time,
Composition etc.
Defective,
2. p, np Attribute Defective rate Nonconformities

No. of breakdowns per

unit time, No. of
3. c, u Attribute Defect rate absentees per day, No.
of defects per unit
area/length/volume etc.
April 2, 2010 17

Control Charts for Variables

• Variable - a single quality characteristic that can be measured on a

numerical
i l scale.
l
• When working with variables, we should monitor both the mean value
of the characteristic and the variability associated with the
characteristic.

April 2, 2010 18

9
 4/2/2010

Control Charts for x and R

Notations
• n - size
i off the
th samplel (subgroup)
( b )
• m - number of samples selected
• x i= average of the observations in the i-th sample (where i = 1, 2, ..., m)
• x = grand average or “average of the averages (this value is used as the
center line of the control chart)
• Ri = range of the values in the ith sample
Ri = xmax - xmin = average range for all m samples
• μ is the true process mean
• σ is the true process standard deviation

April 2, 2010 19

Control Limits for the x chart

UCL = x + A 2 R
Center Line = x
LCL = x − A 2 R
• Control Limits for the R chart
UCL = D4R
C e n t e r L in e = R
LC L = D 3R

• A2 , D3 and D4 are found for various values of n.

Read Table 16.2
April 2, 2010 20

10
 4/2/2010

Control Charts for Individual Measurements

• If you do not get a sample size greater than 1 (n =1)?

– Automated inspection and measurement technology is used,
andd every unit
i manufactured
f d is
i analyzed.
l d
– The production rate is very slow, and it is inconvenient to
allow samples sizes of n > 1 to accumulate before analysis
– Repeat measurements on the process differ only because of
laboratory or analysis error, as in many chemical processes.

• The X and MR charts are useful for samples

p of sizes n = 1.

April 2, 2010 21

Moving Range Chart

• The moving range (MR) is defined as the absolute difference between
two successive observations:
MRi = |xi - xi-1|
which will indicate possible shifts or changes in the process from one
observation to the next.

X and Moving Range Charts

• The X chart is the plot of the individual observations. The control
limits are
MR
UCL = x + 3
d2
CL = x
MR
LCL = x − 3
d2
m
∑ MRi
where MR = i=1
m
April 2, 2010 22

11
 4/2/2010

X and Moving Range Charts

• The control limits on the moving range chart are:

UCL= D4 MR
CL = MR
LCL= 0

April 2, 2010 23

Example
Ten successive heats of a steel alloy are tested for hardness.
Heat Hardness Heat Hardness
1 52 6 52
2 51 7 50
3 54 8 51
4 55 9 58
5 50 10 51
I and MR Chart for hardness

62 3.0SL=60.97
Individuals

52 X=52.40

-3.0SL=43.83
3 0SL=43 83
42
Observation 0 1 2 3 4 5 6 7 8 9 10

10 3.0SL=10.53
Moving Range

5
R=3.222

0 -3.0SL=0.000

April 2, 2010 24

12
 4/2/2010

Control Charts for Attributes

• Data that can be classified into one of several

categories or classifications is known as attribute
data.
• Classifications such as conforming and
nonconforming are commonly used in quality
control.
• Another example of attribute data is the count of
defects.

April 2, 2010 25

Control Charts for Fraction Nonconforming

• Fraction
i nonconforming
f i is i the
h ratio
i off the
h number
b
of nonconforming items in a population to the
total number of items in that population.
• Control charts for fraction nonconforming are
based on the binomial distribution.

April 2, 2010 26

13
 4/2/2010

Development of the Fraction Nonconforming Control Chart

Assume
• n = number of units of product selected at random.
• D = number of nonconforming units from the sample
• p = probability of selecting a nonconforming unit from the sample.
• Then,
⎛n⎞
P(D = x) = ⎜⎜ ⎟⎟p x (1 − p) n − x
⎝x⎠
• The sample fraction nonconforming is given as
D
p̂ =
n
where p̂ is a random variable with mean and variance
p (1 − p )
μ=p σ2 =
n

April 2, 2010 27

Standard Given
• If a standard value of p is given, then the control limits for the fraction
nonconforming are
p(1− p)
UCL = p + 3
n
CL = p
p(1− p)
LCL = p − 3
n
No Standard Given
• If no standard value of p is given, then the control limits for the fraction
nonconforming are
p (1 − p )
UCL = p + 3
n
CL = p
p (1 − p )
LCL = p − 3
n
where m m
∑ D i ∑ p̂ i
p = i=1
= i=1
mn m

April 2, 2010 28

14
 4/2/2010

Example
• A process that produces bearing housings is investigated. Ten samples
of size 100 are selected.
Sample # 1 2 3 4 5 6 7 8 9 10
# Nonconf. 5 2 3 8 4 1 2 6 3 4
• I this
Is hi process operating
i ini statistical
i i l control?
l?

n = 100, m = 10
Sample # 1 2 3 4 5 6 7 8 9 10
# Nonconf. 5 2 3 8 4 1 2 6 3 4
Fraction
0.05 0.02 0.03 0.08 0.04 0.01 0.02 0.06 0.03 0.04
Nonconf.

m
∑ p̂ i
p = i=1
= 0 . 038
m

April 2, 2010 29

Example
Control Limits are: 0.038(1 − 0.038)
UCL = 0.038 + 3 = 0.095
100
CL = 0.038
0.038(1 − 0.038)
LCL = 0.038 − 3 = −0.02 → 0
100

P Chart for C1
0.10
3.0SL=0.09536
Proportion

0.05
P=0 03800
P=0.03800
P

0.00 - 3.0SL=0.000

0 1 2 3 4 5 6 7 8 9 10
Sampl e Number

April 2, 2010 30

15
 4/2/2010

The np control chart

• The actual number of nonconforming can also be
charted. Let n = sample size, p = proportion of
nonconforming.
f i The
h controll limits
li i are:
UCL = np + 3 np (1 − p )
CL = np
LCL = np − 3 np (1 − p )
(if a standard, p, is not given, use p )

April 2, 2010 31

Control Chart (Attribute) with Variable

Sample Size
• In some applications of the control chart for the fraction
nonconforming, the sample is a 100% inspection of the process
output
t t over some period
i d off time.
ti
• Since different numbers of units could be produced in each
period, the control chart would then have a variable sample size.

Three Approaches for Control Charts

1. Variable Width Control Limits

2
2. C t l Limits
Control Li it Based
B d on Average
A Sample
S l Size
Si
3. Standardized Control Chart

April 2, 2010 32

16
 4/2/2010

Variable Width Control Limits

• Determine control limits for each individual
sample that are based on the specific sample
size.
size
• The upper and lower control limits are

p (1 − p )
p±3
ni

April 2, 2010 33

Control Limits Based on an Average Sample Size

• Control charts based on the average sample size
results in an approximate set of control limits.
• The average sample size is given by
m
∑ ni
n= i =1
m
• The upper and lower control limits are
p (1 − p )
p±3
n

April 2, 2010 34

17
 4/2/2010

Control Chart (Attribute) with Constant

Sample Size
c-chart (Control Chart for Total Nonconforming)
• Standard Given:
Gi en: UCL = c + 3 c
CL = c
LCL = c − 3 c

• N St
No Standard
d d Gi
Given: UCL = c + 3 c
CL = c
LCL = c − 3 c

April 2, 2010 35

Choice of Sample Size: The u Chart

• If we find c total nonconformities in a sample of n
inspection units, then the average number of
nonconformities per inspection unit is u = c/n.
c/n
• The control limits for the average number of
nonconformities is
u
UCL = u + 3
n
CL = u
u
LCL = u − 3
n

April 2, 2010 36

18
 4/2/2010

Control Chart (Attribute) with Variable

Sample Size
Three Approaches

1. Variable Width Control Limits

2. Control Limits Based on Average Sample Size
3. Standardized Control Chart

April 2, 2010 37

Variable Width Control Limits

• Determine control limits for each individual
sample that are based on the specific sample
size.
size
• The upper and lower control limits are

u
u±3
ni

April 2, 2010 38

19
 4/2/2010

Control Limits Based on an Average Sample Size

• Control charts based on the average sample size
results in an approximate set of control limits.
• Theh average sample l size
i isi given
i by
b
m
∑ ni
n= i =1
m
• The upper and lower control limits are
u
u±3
n

April 2, 2010 39

The Standardized Control Chart

• The points plotted are in terms of standard
deviation units. The standardized control chart
has the follow properties:
– Centerline at 0
– UCL = 3 LCL = -3
– The points plotted are given by:
ui − u
zi
u
ni

April 2, 2010 40

20
 4/2/2010

Choice Between Attributes and Variables Control Charts

• Each has its own advantages and disadvantages

• Attributes data is easy to collect and several
characteristics
h i i may be b collected
ll d per unit.
i
• Variables data can be more informative since specific
information about the process mean and variance is
obtained directly.
• Variables control charts provide an indication of
impending trouble (corrective action may be taken
before any defectives are produced).
produced)
• Attributes control charts will not react unless the process
has already changed (more nonconforming items may be
produced.

April 2, 2010 41

Guidelines for Implementing Control Charts

1 Determine which process characteristics to

1.
control.
2. Determine where the charts should be
implemented in the process.
3. Choose the proper type of control chart.
4. Take action to improve
p processes as the result of
p
SPC/control chart analysis.
5. Select data-collection systems and computer
software.

April 2, 2010 42

21