1

Chapter 10
Quality Control
2
Phases of Quality Assurance
Acceptance
sampling
Process
control
Continuous
improvement
Inspection
before/after
production
Corrective
action during
production
Quality built
into the
process
The least
progressive
The most
progressive
3
Inspection: Appraisal of good/service quality
How Much (sample size) /How Often (hourly, daily)
C
o
s
t

Optimal
Amount of Inspection
Cost of
inspection
(appraisal and
Prevention cost)
Cost of
passing
defectives
(failure cost)
Total Cost
4
Inspection
Where/When
Raw materials
Finished products







Before a costly operation, PhD comp. exam before candidacy
Before an irreversible process, firing pottery
Before a covering process, painting, assembly
Centralized vs. On-Site, my friend checks quality at cruise lines
Inputs Transformation Outputs
Acceptance
sampling
Process
control
Acceptance
sampling
5
Examples of Inspection Points
Type of
business
Inspection
points
Characteristics
Fast Food Cashier
Counter area
Eating area
Building
Kitchen
Accuracy
Appearance, productivity
Cleanliness
Appearance
Health regulations
Hotel/motel Parking lot
Accounting
Building
Main desk
Safe, well lighted
Accuracy, timeliness
Appearance, safety
Waiting times
Supermarket Cashiers
Deliveries
Accuracy, courtesy
Quality, quantity
6
Statistical Process Control (SPC)
SPC: Statistical evaluation
of the output of a process during production

The Control Process
Define
Measure
Compare to a standard
Evaluate
Take corrective action
Evaluate corrective action
7
Statistical Process Control
Shewharts classification of variability:
common cause vs. assignable cause
Variations and Control
Random variation: Natural variations in the
output of process, created by countless minor
factors, e.g. temperature, humidity variations.
Assignable variation: A variation whose source
can be identified. This source is generally a
major factor, e.g. tool failure.
8
Mean and Variance
Given a population of numbers, how to
compute the mean and the variance?
o

= =
= =
=

=
=
deviation Standard
) (
Variance
Mean
} ,..., , { Population
1
2
2
1
2 1
N
x
N
x
x x x
N
i
i
N
i
i
N
9
Statistical Process Control
From a large population of goods or
services (random if possible) a sample is
drawn.
Example sample: Midterm grades of BA3352
students whose last name starts with letter R
{60, 64, 72, 86}, with letter S {54, 60}
Sample size= n
Sample average or sample mean=
Sample range= R
Standard deviation of sample means=




x
population the of deviation Standard : where o
o
o
n
x
=
10
Sampling Distribution
Sampling distribution
Variability of the average scores of
people with last name R and S
Process distribution
Variability of the scores
for the entire class
Mean
Sampling distribution is the distribution of sample means.
Grouping reduces the variability.
11
Normal Distribution
Mean
3o 2o +2o +3o
95.44%
99.74%
x
at x. cdf normal ) 1 , _ , , ( normdist : functions l statistica Excel
at x. pdf normal ) 0 , _ , , ( normdist : functions l statistica Excel
dev st mean x
dev st mean x
normdist(x,.,.,0)
Probab
normdist(x,.,.,1)
12
Cumulative Normal Density
) _ , , ( norminv : prob" " at cdf of function Inverse
) 1 , _ , , ( normdist : at x (cdf) function Cumulative
: functions l statistica Excel
dev st mean prob
dev st mean x
0
1
x
normdist(x,mean,st_dev,1)
prob
norminv(prob,mean,st_dev)
13
Normal Probabilities: Example
If temperature inside a firing oven has a
normal distribution with mean 200
o
C and
standard deviation of 40
o
C, what is the
probability that
The temperature is lower than 220
o
C
=normdist(220,200,40,1)
The temperature is between 190
o
C and 220
o
C
=normdist(220,200,40,1)-normdist(190,200,40,1)
14
Control Limits
Sampling
distribution
Process
distribution
Mean
LCL
Lower
control
limit
UCL
Upper
control
limit
Process is in control if sample mean is between control limits.
These limits have nothing to do with product specifications!
15
Setting Control Limits:
Hypothesis Testing Framework
Null hypothesis: Process is in control
Alternative hypothesis: Process is out of control
Alpha=P(Type I error)=P(reject the null when it is true)=
P(out of control when in control)
Beta=P(Type II error)=P(accept the null when it is false)
P(in control when out of control)

If LCL decreases and UCL increases what happens to
Alpha ?
Beta?
Not possible to target alpha and beta simultaneously,
control charts target a desired level of Alpha.

16
Type I Error=Alpha
Mean
LCL UCL
o/2 o/2
o = Probability
of Type I error
st_dev) mean, /2, - norminv(1 UCL
st_dev) mean, /2, norminv( LCL
o
o
=
=
17
Control Chart
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
UCL
LCL
Sample number
Mean
Out of
control
Normal variation
due to chance
Abnormal variation
due to assignable sources
Abnormal variation
due to assignable sources
18
Observations from Sample Distribution
Sample number
UCL
LCL
1 2 3 4
19
Control Charts
Control charts for variables (measurable
quantities), e.g. length, temperature
Mean control charts
To check mean
Range control charts
To check variability
Control charts for attributes, e.g. fit, defective
p-charts
To check proportion of defectives (occurrences)
c-charts
To check the number of defectives (occurrences)
20
Mean control chart
deviation standard of multiple a minus mean grand
deviation standard of multiple a plus mean grand

= =
= + =
=
x
x
z x LCL
z x UCL
x of average x mean Grand
o
o
x
x
x
) , x /2, - norminv(1
z
o
o o
o
x x UCL
=

=
Most often z is set to 2 or 3.
If the standard deviation of the sample means is not known,
use the average of sample ranges to get the limits:
ranges sample of average the of multiple a minus mean grand
ranges sample of average the of multiple a plus mean grand
R ranges sample of average
2
2
= =
= + =
=
R A x LCL
R A x UCL
R
Multiplier A_2 depends on n and is available in Table 10-2.
21
Range Control Chart
ranges sample of average the of multiple A
ranges sample of average the of multiple A
3
4
= =
= =
R D LCL
R D UCL
Multipliers D_4 and D_3 depend on n and are available in Table 10-2.
EX: In the last five years, the range of GMAT scores of incoming PhD class is
88, 64, 102, 70, 74. If each class has 6 students, what are UCL and LCL for
GMAT ranges?
0 79.6 * 0 159.2 79.6 * 2
. 0 D , 2 D 6, n For . 6 . 79 5 / ) 74 70 102 64 88 (
3 4
3 4
= = = = = =
= = = = + + + + =
R D LCL R D UCL
R
Are the GMAT ranges in control?
22
Mean and Range Charts: Which?
UCL
LCL
UCL
LCL
R-chart
x-Chart
Detects shift
Does not
detect shift
(process mean is
shifting upward)
Sampling
Distribution
23
Mean and Range Charts: Which?
UCL
LC
L
LC
L
R-chart
Reveals increase
x-Chart
UCL
Does not
reveal increase
(process variability is increasing)
Sampling
Distribution
24
Use of p-Charts
p=proportion defective, assumed to be known
When observations can be placed into two categories.
Good or bad
Pass or fail
Operate or dont operate
Go or no-go gauge





before as ,
) 1 (

LCL
z
n
p p
where
z p z p UCL
p
p p

=
= + =
o
o o
25
Use of c-Charts
c=number of occurrences per unit
Use only when the number of occurrences per
unit can be counted.
Scratches, chips, dents, or errors per item
Cracks or faults per unit of distance
Breaks or Tears per unit of area
Bacteria or pollutants per unit of volume
Calls, complaints, failures per unit of time
c average the use known, not is c if
LCL c z c c z c UCL = + =
26
C-chart Example
While the nuclear submarine Kursk was being raised in the
Barents sea (between Svalbard, No and Novaya Zemlya, Ru),
which took 15 hours, engineers took a reading of number of
Geiger counts per hour to detect any increase in radiation
levels. Should they have stopped before 5
th
or 10
th
hour given
3-sigma control and the readings data: 42, 48, 50, 45, 52, 66,
64, 84, 92, 76.
At the 5
th
hour, average number of counts=47.4, stdev of counts=6.88,
UCL=47.4+3*6.88=68.05, LCL=47.4-3*6.88=26.75. Do not stop.
At the 10
th
hour, average number of counts=61.9, stdev of counts=7.87,
UCL=61.9+3*7.87=85.51, LCL=61.9-3*7.87=38.29. Stop, 9
th
reading is
out of control.
27
Up and Down Run Charts
If all readings are in control, is the process
really in control?
There could be trends in readings even
when they are in control.

Counting Up/Down Runs (r=8 runs)
U U D U D U D U U D
28
Up and Down Run Charts
samples of Number K
90
29 16
and
3
1 - 2K
E(r)
runs of stdev of multiple a minus runs Expected ) (
runs of stdev of multiple a plus runs Expected ) (
=

= =
= =
= + =
K
z r E LCL
z r E UCL
r
r
r
o
o
o
EX: What are 3-sigma UCL and LCL for the number of runs in 50 samples?
2.92 * 3 - 33 ) (
2.92 * 3 33 ) (
92 . 2
90
29 16
and 33
3
1 - 2K
E(r) 50, K
= =
+ = + =
=

= = = =
r
r
r
z r E LCL
z r E UCL
K
o
o
o
29
Tolerances/Specifications
Requirements of the design or customers
Process variability
Natural variability in a process
Variance of the measurements coming from the process
Process capability
Process variability relative to specification
Capability=Process specifications / Process variability

Process Capability
30
Process Capability:
Specification limits are not control chart limits
Lower
Specification
Upper
Specification
Process variability matches
specifications
Lower
Specification
Upper
Specification
Process variability well within
specifications
Lower
Specification
Upper
Specification
Process variability exceeds
specifications
Sampling
Distribution
is used
31
Process Capability Ratio
When the process is centered, process capability ratio
Upper specification lower specification
6o
Cp =
A capable process has large Cp.

Example: The standard deviation, of sample averages of the
midterm 1scores obtained by students whose last names start
with R, has been 7. The SOM management requires the
scores not to differ by more than 50% in an exam. That is the
highest score can be at most 50 points above the lowest
score. Suppose that the scores are centered, what is the
process capability ratio?
Answer: 50/42
32
Process Capability Ratio
When the process is not centered, process capability ratio
Min{Process mean - lower spec , Upper spec - Process mean}
3o
Cpk=
When the process is not centered, the closest spec to mean determines
the capability of the process because that spec is likely to be
more of a limiting factor than the other.

Example: Suppose that the process is not centered in the previous example
and the SOM wants all the scores to fall within 50% and 100%. What is the
Capability ratio if the average score was 70?

Answer: From the lower limit, we have (70-50)/21
From the upper limit, we have (100-70)/21
Then the ratio is 20/21
33
Process
mean
Lower
specification
Upper
specification
+/- 3 Sigma
+/- 6 Sigma
3 Sigma and 6 Sigma Quality
34
Chapter 10 Supplement
Acceptance
Sampling
35
Acceptance Sampling
Acceptance sampling: Is a lot of N products good
if a random sample of n (n<N) products contain
only c defects?
For example take a sample of 10(=n) milk bottles out
of every 100(=N). If 1(=c) or more bottles do not fit
specifications, reject the entire lot of 100 bottles.
c is determined to balance type I and type II
errors.
This is a smart compromise between 100%
inspection and no inspection.
Generally used for input/output inspection.

36
Why not to emphasize
Acceptance Sampling (AS)
AS plans have no clearly stated economic objective.
They target some levels of type I and II errors.
AS incorporate an attitude of punishment by
rejecting entire lots after examining small samples.
This feeds the mistrust between supplier and the
customer.
AS does not attempt to find the root cause of
defectives. It merely detects defectives. Real
problem is actually finding the root cause. Some
people say that:
AS provides elegant solutions to balance type I and II
errors by making a type III error: solving the wrong
problem.