Chapter 6: Introduction To Analysis of Variance, Statistical Quality Control and System Reliability
Chapter 6: Introduction To Analysis of Variance, Statistical Quality Control and System Reliability
Chapter 6: Introduction To Analysis of Variance, Statistical Quality Control and System Reliability
Introduction
Analysis of variance (ANOVA) is a procedure used to test the null hypothesis that the means of 3 or more
populations are equal.
Example 6.1. ABC manufactures printers. It has 3 plants, Plant A, Plant B and Plant C. To measure how
much employees at these plants know about the safety procedures, a random sample of 6 employees was
selected from each plant and given an examination. The examination scores of these 18 employees are
obtained. The managers can use these data to test the hypothesis that the mean examination score is the
same for all three plants via performing an ANOVA.
The two variables in this example are the 3 plants names, Plant A, Plant B and Plant C, and the
examination scores. Because the objective is to determine whether the mean examination score is the
same for the three plants, examination score is referred to as the dependent or response variable and plant
name as the independent variable or factor. In general, the values of a factor selected for investigation
are referred to as levels of the factor or treatments. Thus, in this example, the three treatments are Plant A,
Plant B and Plant C. These three treatments define the populations of interest. For each treatment or
population, the response variable is the examination score.
6.2
In one way ANOVA, we analyze only one factor. A completely randomized design experiment can be
analyzed by a one way ANOVA.
Assumptions for One Way ANOVA
1.
The populations from which the samples are drawn are (approximately) normally distributed.
2.
The populations from which the samples are drawn have the same variance 2 .
3.
The samples drawn from different populations are random and independent.
2.
3.
4.
Chapter 6 - 1
Sum of Squares
Treatments,
SSTR
Error, SSE
Defining formula
k
SSTR =
SSE =
n j (x j x )
j =1
k
nj
j =1 i =1
Total, SST
SST =
nj
j =1 i =1
Note
1.
2.
3.
Computing formula
(x
xj )
SSE = x 2 T j2 n j
x)
ij
(x
ij
) (nx )
SSTR = T j2 n j
Mean Square
SSTR
MSTR =
k 1
MSE =
SSE
nk
Sum of
Squares
SSTR
SSE
SST
Degrees of
Freedom
k 1
nk
n 1
Mean Square
MSTR
MSE
Example 6.2. 15 students were randomly assigned to 3 groups to experiment with 3 different methods of
teaching arithmetic. At the end of the semester, the same test was given to all 15 students. The following
table (in the solution) gives the scores of students in the 3 groups
At the 1% significance level, can we reject the null hypothesis that the mean arithmetic score of all
students taught by each of these 3 methods is the same? Assume that all the required assumptions for one
way ANOVA procedure hold true.
Chapter 6 - 2
Solution.
Let 1 , 2 and 3 be the mean arithmetic scores of all the students taught by Method I, Method II and
Method III, respectively.
Method I
48
73
51
65
87
T1 =
n1 =
Method II
55
85
70
69
90
T2 = 369
n2 =
Method III
84
68
95
74
67
T3 = 388
n3 =
x = T1 + T2 + T3 =
x 2 =
) (nx )
SSTR = T j2 n j
MSTR =
SSTR
=
k 1
SSE = x 2 T j2 n j =
MSE =
SSE
=
nk
Degrees of
Freedom
Chapter 6 - 3
Mean Square
Example 6.3. The research department of a bank wanted to check whether the 4 tellers at a branch of
this bank serve, on average, the same number of customers per hour. The research manager observed
each of the 4 tellers for a certain number of hours. The following table (in solution) gives the number of
customers served by the 4 tellers during each of the observed hours
At the 5% significance level, test the null hypothesis that the mean number of customers served per hour
by each of these 4 tellers is the same. Assume that all the assumptions required to apply the one way
ANOVA procedure hold true.
Solution.
Let 1 , 2 , 3 and 4 be the mean number of customers served per hour by the tellers A, B, C and D,
respectively.
Teller A
19
21
26
24
18
Teller B
14
16
14
13
17
13
T2 = 87
n2 =
T1 =
n1 =
x = 398
Sum of
Squares
Teller C
11
14
21
13
16
18
T3 = 93
n3 =
Teller D
24
19
21
26
20
T4 = 110
n4 =
x 2 = 7614
n = 22
Degrees of
Freedom
Mean Square
Chapter 6 - 4
6.3
Notation
xij = value of the observation corresponding to treatment j in block i .
Blocks
Controller 1
System A
15
Treatments
System B
15
System C
18
Block Means
x1 = 16.0
Controller 2
14
14
14
x 2 = 14.0
Controller 3
10
11
15
x3 = 12.0
Controller 4
13
12
17
x 4 = 14.0
Controller 5
16
13
16
x5 = 15.0
Controller 6
13
13
13
x6 = 13.0
x 1 = 13.5
x 2 = 13.0
x 3 = 15.5
x = 14.0
Treatment Means
SST =
j =1 i =1
SSTR =
k
j =1
SSBL =
b
i =1
(x
x) =
2
ij
b (x j x ) =
2
k (xi x ) =
2
Sum of
Squares
Degrees of
Freedom
Mean Square
Chapter 6 - 5
6.4
Control Charts
Control chart
A control chart consist of:
Upper Control Limit (UCL),
Lower Control Limit (LCL),
Center Line (CL),
Plot of quality characteristic measured form a sample versus sample number or time,
Sample points are connected by straight lines.
UCL
CL
LCL
6.5
i.
ii.
Quality characteristics that can be expressed as numerical measures are called variables. For examples,
diameter of a piston, weight or volume.
For quality characteristics that is a variable, control charts used are:
X -chart to control process average/mean.
R-chart for range
s-chart for standard deviation
It is important to control over both the process mean and process variability.
Control charts for X and R
Suppose X N ( , 2 ) , m samples, each containing n observations are taken.
Case 1 :
02
n
) , E ( X i ) = 0 and X i =
Chapter 6 - 6
0
n
UCL = 0 + z 2
and
LCL = 0 z 2
0
n
is 1 .
LCL = R + 3 R = D1 0
CL = R = d 2 0
UCL = R 3 R = D 2 0
and unknown
Case 2 :
Estimate and from preliminary samples of a process that is in control.
Take m > 20 , n = 4, 5, 6 .
For each sample i, sample mean X i is computed.
The grand average, X =
X 1 + X 2 + ... + X m
is the best estimator of the process average, ..
m
LCL = 3
R1 + R2 + ... + Rm
is the best estimator for R .
m
= X A2 R
CL = = X
UCL = + 3
= X + A2 R
For R-chart:
LCL = R 3 R = D3 R
CL = R = R
UCL = R + 3 R = D4 R
Chapter 6 - 7
Note :
1.
An R-chart is usually analyzed before the X -chart to determine out-of-control situations. R-chart
reflects process variability must be in control for a meaningful X -chart.
2.
If the X -chart shows that process mean also in control, then we conclude that the process is in
control.
3.
If the process variability is out-of-control, investigation is done to check for assignable causes.
These out-of-control point(s) will be deleted and the remaining samples are used to determine the
revised control limits.
Example 6.5. The data below are X and R values for 12 samples of size n = 5 taken from a process.
Assume the process output is normally distributed.
Sample number
1
2
3
4
5
6
7
8
9
10
11
12
X
35.4
34.0
37.1
34.9
33.5
31.7
34.0
35.1
33.7
32.8
33.5
34.2
R
8
6
5
7
4
3
8
4
2
1
3
2
Set up the 3-sigma control limits for X and R charts on this process. Does the process seem to be in
statistical control? If necessary, revise the trial control limits.
Solution.
Chapter 6 - 8
Chapter 6 - 9
6.6
Control charts for attributes monitor a process for qualitative attributes (being defective / nonconforming)
p-chart : The control chart for fraction nonconforming
If an item does not conform to standard on one or more of the several quality characteristics that are
examined by inspector, the item is classified as nonconforming.
Given standard value of p ( p = p 0 )
Case 1 :
LCL = p 0 3
p 0 (1 p 0 )
n
CL = p 0
UCL = p 0 + 3
p 0 (1 p 0 )
n
Case 2 :
Note : 1)
2)
If LCL < 0 , then we set LCL = 0 , since p 1 , p 2 , ... p n are positive values.
Points below LCL may not indicate an improved quality. Often they are caused by errors,
inexperienced inspectors, improper calibrated test instrument.
Example 6.6. In each of 13 consecutive and recent years, 100,000 subjects were randomly selected and
the number who died from respiratory tract infections is recorded, with the results given here. Construct a
control chart for p and determine whether the process is within statistical control.
Numbers of
deaths
25
30
24
33
22
32
25
33
27
32
Solution.
Chapter 6 - 10
30
31
31
6.7
System Reliability
Most products are made up of a number of components. The reliability of each component and the
configuration of the system consisting of these components determine the system reliability.
Example 6.7. A small series system consists of 4 components with reliability of 0.98, 0.97, 0.95 and
0.99. Determine the reliability of the system.
Solution.
Chapter 6 - 11
Example 6.8. A missile system has 4000 components. The reliability of each component is extremely
high with Ri = 0.9999 for i = 1, 2, ..., 4000 . Find the probability that the missile will operate. If the
components can be halved, what will be the reliability of the system?
Example 6.9. If a typewriter consists of 1000 components, all of which must operate, each with identical
reliability, what should be the reliability of each component so that the typewriter has a reliability of
0.99?
1
2
.
n
Suppose we have n components in parallel with reliability of the ith component denoted by Ri ,
i = 1, 2, ..., n . Assuming the components operate randomly and independently, the probability of failure
of each component Fi = 1 Ri .
The system fails only if all the components fail.
n
i =1
i =1
unreliability of a system) Fs = Fi = (1 Ri ) .
n
Chapter 6 - 12
Example 6.10. Consider a system consisting of three components in parallel, each with a reliability of
0.65. Determine the reliability of the system.
Example 6.11. Determine the reliability of the system shown, which consists of 8 components with the
following reliability. Find the reliability of the system.
A1 = A2 = A3 = 0.72 ; B = 0.96 ; C1 = C 2 = 0.87 ; C 3 = C 4 = 0.91 .
A1
A2
C1
C3
C2
C4
A3
Chapter 6 - 13
Chapter 6 - 14