1

INSE 6220 -- Week 6

Advanced Statistical Approaches to Quality

Cumulative Sum (CUSUM) Control Charts

Exponentially Weighted Moving Average
(EWMA) Control Charts
Midterm Review

Dr. A. Ben Hamza Concordia University

 2

The Cumulative Sum (CUSUM) Control Chart

Although Shewhart charts with 3 limits can quickly detect large
process changes, they are ineffective for small, sustained process
changes (for example, changes smaller than 1.5 )
Two alternative control charts have been developed to detect small
changes: the CUSUM and EWMA control charts.
They also can detect large process changes (for example, 3 shifts),
but detection is usually somewhat slower than for Shewhart charts.
CUSUM Control Chart
The cumulative sum (CUSUM) is defined to be a running summation of
the deviations of the plotted variable from its target.
i
If the sample mean is plotted, the cumulative sum is Ci ( x j 0 )
j 1

where 0 is the target for x

 3

Interpretation of the CUSUM chart

i i 1
C i ( x j 0 ) ( x j 0 ) ( x i 0 ) C i 1 ( x i 0 )
j1 j1
=0, Ci is a random walk with mean zero
>0, Ci is an upward drift trend
<0, Ci is a downward drift trend

Remark: a trend of Ci is an indication of the process mean shift.

 4

Using CUSUM

i
Ci (x
j 1
j 10)
The Tabular CUSUM for Monitoring the Process Mean
5

Ci max 0, xi ( 0 K ) Ci1
Ci max 0, ( 0 K ) xi Ci1
where the starting values are : C0 C0 0

| 1 0 |
K k
2 2

K is usually called the reference value

A deviation from the target that is larger than K increases either C+ or C-
A control limit violation occurs when either C+ or C- exceeds a specified control limit (or
threshold) H (e.g. H=5).
After a limit violation occurs, that sum is reset to zero or to a specified value.
Ci and Ci are accumulation of deviations above and below the target mean
Ci is called: one-sided upper cusum
Ci is called: one-sided lower cusum
cusum control chart plots the values of Ci and Ci for each sample
 6

The Standardized CUSUM

Recommendations for CUSUM Design: k 0.5 and h 5

 7

Exponentially Weighted Moving Average (EWMA) Control Chart

zi xi (1 ) zi 1

0 1

z0 0

z2 2 1 (1 ) 2i

i
2
i 1
zi (1 ) j xi j (1 )i z0
j 0
 8

Exponentially Weighted Moving Average (EWMA) Control Chart

UCL 0 L 1 (1 ) 2i
2
CL 0

LCL 0 L 1 (1 ) 2i
2

(1 ) 2i 0, when i
Steady-state
control limits 10.8
Exponentially Weighted Moving Average (EWMA) Chart

10.7
27
UCL 0 L 26
2
10.6
UCL
10.5

LCL 0 L EWMA 10.4

2 10.3 CL
10.2

10.1

10 LCL
L = width of control limits (given) 3
9.9
0 5 10 15 20 25 30
Sample Number
9
 10

Period 1 Control Limits: Steady-State Control Limits:

UCL
 11

Main Points for Midterm Exam

What to Study
Some topics are more important than others.
Spend your time on the right stuff.
Dont waste time on topics we havent emphasized in class.
How to Prepare for the Midterm
Focus on the main topics
Go back to your assignment. There is a lot of good review in there.
Make a list of your problem areas.
Eliminate any problems not mentioned on the lecture slides
Keep the class notes as guideline.
Work on relevant exercises at the end of the chapters.
Bring a Calculator on the day of the exam

Midterm Exam Coverage: Lectures 1-to-5, and Assignment 1

 12

Sometimes called a probability Sometimes called a probability

mass function density function
 13

Mean of a Random Variable

 14

Variance of a Random Variable

The (population) variance of random variable (RV) gives an idea of how
widely spread the values of the RV are likely to be. It is the second moment of
the distribution, indicating how closely concentrated around the expected
value of the distribution is. The variance is defined by

Var ( X ) E ( X 2 ) ( E ( X ))2 2

Var ( X ) is referred to as the standard deviation

Example
 15

Sample mean and sample variance

Mean

variance

s s2
 16

Descriptive Statistics
 17

Graphical Representation of Data

 18

Binomial density function

The discrete binomial pdf
n!
f ( x) p x (1 p ) (n x )
x !( n x )!
assigns probability to the event of x successes in n trials of a Bernoulli process
(such as coin flipping) with probability p of success at each trial.

Expected Value
E(X) = = np
Variance
Var(X) = 2 = np(1 - p)
 19

Poisson Probability Distribution

The Poisson distribution is
e x
f ( x) x 0,1, 2,...
x!
Where the parameter >0 is the mean number of successes in the interval.
The mean and variance of the Poisson distribution are

and 2
 20

Uniform Distribution
 21

Normal Probability Distribution

The normal probability distribution is the The normal distribution is
most important distribution for describing 1
( x )2

f ( x) e 2 2
x
a continuous random variable. 2
It has been used in a wide variety of with mean and variance 2
applications: The normal distribution is: X N ( , 2 )
Heights and weights of people
The visual appearance of the normal
Test scores distribution is a symmetric, unimodal or
Scientific measurements bell-shaped curve as shown in the figure.
Amounts of rainfall
It is widely used in statistical inference
 22

Calculating Normal Probabilities

P(45 < X < 60) ?
mean of 50 minutes and a
standard deviation of 10 minutes

0
 23

Normal Approximation to the

Binomial Distributions

Let X be a binomial rv based on n trials, each

with probability of success p. If the binomial
probability histogram is not too skewed, X may be
approximated by a normal distribution with

np and np (1 p )

x 0.5 np
P( X x)
np (1 p )

 24

Exponential Distribution
A continuous random variable X has an
exponential distribution with parameter if
the probability density function (pdf) is
e x x0
f ( x; )
0 otherwise
The mean and variance of a random variable X having the
exponential distribution

1 2 1

2
 25

Central Limit Theorem

 26

Chi-squared distribution
 27

Sampling: the t-student distribution

If Z N (0,1) then Z t with Y k is distributed as a student
Y /k
k
2

t with k degrees of freedom.

Typical use: Find distribution of average when is NOT known

For k , tk N (0,1)

Consider X i N ( , 2 ) Then
X
X / n N (0,1)
t n 1
s/ n s / 1
n21
n 1
This is just the normalized distance from mean (normalized to our
estimate of the sample variance).
 28

Students t-distribution
 29

F distribution
 30

Summary of Control Charts

X LCL x A2 R LCL D3 R
Xbar & R chart CL x CL R
R

d2 UCL x A2 R UCL D4 R

X LCL x A3 S LCL B3S

Xbar & S chart CL x CL S
S
UCL x A3S UCL B4 S
c4

If X 1 , X 2 ,..., X n is a random sample from a N ( , 2 ) population, then

E (S 2 ) 2 but E (S )
 31

Example

x Chart :
UCL x A2 R
1.5056 (0.577)(0.32521) 1.69325
Central line x 1.5056
LCL x A2 R
1.5056 (0.577)(0.32521) 1.31795

R Chart :
UCL D4 R (2.114)(0.32521)
Central line R 0.32521
LCL D3 R (0)(0.32521)

x 1.50345
R 0.3360
 32

Example
The grand average and the
average sample standard
deviation are:
1 25
X X i 74.001
25 i 1
1 25
S Si 0.0094
25 i 1

X chart
UCL X A3 S
74.001 (1.427)(0.0094) 74.014
CL X 74.001
LCL X A3 S
74.001 (1.427)(0.0094) 73.988

S chart
UCL B4 S
(2.089)(0.0094) 0.0196
CL S 0.0094
LCL B3 S
(0)(0.0094) 0
33
 34

Process capability analysis

Capability indices:
Process capability potential
Cp = (USL LSL) / 6x

Upper capability index

CpU = (USL X ) / 3x

Lower capability index

CpL = ( X LSL) / 3x

Process capability index

Cpk = min (CpU, CpL)
 35

Process capability analysis (cont.)

Estimating Process Capability

The X-bar and R charts give information about the capability of the
process relative to its specification limits.
We can estimate the fraction of nonconforming items for any
process where specification limits are involved.
Assume the process is normally distributed, and X is normally
distributed, then the fraction nonconforming can be found by
calculating:
 36

Inference on the mean of a population, variance known

H 0 : 0
H1 : 0 (3-22)

X 0
Z0 (3-23)
/ n

H1 in equation (3-22) is a two-sided alternative hypothesis

The procedure for testing this hypothesis is to:
take a random sample of n observations on the random variable x,
compute the test statistic, and
reject H0 if |Z0| > Z/2, where Z/2 is the upper /2 percentage of the
standard normal distribution.
In some situations we may wish to reject H0 only if the true mean is larger
than 0
Thus, the one-sided alternative hypothesis is H1: >0, and we would reject
H0: =0 only if Z0>Z
If rejection is desired only when <0
Then the alternative hypothesis is H1: <0, and we reject H0 only if Z0<Z
 37

Inference on the mean of a population with known variance

H 0 : 0
H1 : 0

X 0
Z0
/ n

X z / 2 X z / 2
n n

The p-value is the probability of observing the given sample result under the assumption that the null
hypothesis is true. If the p-value is less than , then you reject the null hypothesis. For example, if =
0.05 and the p-value is 0.03, then you reject the null hypothesis. The converse is not true. If the p-value
is greater than , you have insufficient evidence to reject the null hypothesis.

2[1 | Z 0 |] for a two-tailed test : H 0 : 0 H1 : 0

p -value 1 ( Z 0 ) for an upper-tailed test : H 0 : 0 H1 : 0
( Z ) for a lower-tailed test : H 0 : 0 H1 : 0
0
 38

Inference on the mean of a normal distribution with unknown variance

H 0 : 0
H1 : 0

X 0
t0
s/ n
 39

Confidence Interval on the Mean of a Normal Distribution, Small Sample Size

Summary ofProcedure
Data Distribution:Normali.e.X~ N(, 2)
Population parameters:
Mean,:unknowntobeestimated
Variance,2:unknownes mateusingS2
Sample Size:small(i.e.n<40)
Sample Statistic:SampleMean
Sampling Distribution:Studentt Distribution
ConfidenceIntervals(CIs):
2sidedCI Upper1sidedCI Lower1sidedCI

X t / 2 , n 1
S S S
X t ,n 1 X t ,n 1
n n n
 40

Confidence Interval on the Variance of a Normal Distribution

Summary of Procedure
Data Distribution: Normal i.e. X~ N(, 2)
Population parameters:
Variance, 2: unknown to be estimated
Sample Size: any
Sample Statistic: Sample variance
Sampling Distribution: Chi-squared Distribution
Because chi-square is asymmetric, confidence intervals bounds not symmetric.
Confidence Intervals (CIs):

2sidedCI Upper1sidedCI Lower1sidedCI

( n 1) S 2 ( n 1) S 2
(n 1) S 2
(n 1) S 2

2

/ 2, n 1
2
2
1 / 2 , n 1 2
1 , n 1
,n 1
2
 41

Inference on a population proportion

Hypothesis Testing H 0 : p p0
H1 : p p0

( x 0.5) np0
np (1 p ) if x np0
0 0
Z0
( x 0.5) np0 if x np0
np0 (1 p0 )

The null hypothesis H 0 : p p0 is rejected if | Z 0 | z / 2

 42

Inference on population proportion

H 0 : p p0
Hypothesis Testing:
H1 : p p0

( x 0.5) np0
np (1 p ) if x np0
0 0
Z0
( x 0.5) np0 if x np0
np0 (1 p0 )

The null hypothesis H 0 : p p0 is rejected if | Z 0 | z / 2

Confidence Interval:

p (1 p ) p (1 p )
p z
p p z / 2
/2
n n
 43

Examples:

(0.76) 0.776373
(1.3) ?
( 3) 1 (3) ?
(3.86) ?
 44

t-distribution table
The shaded are is equal
to for t t ,
= degree of freedom

Example:

n 16, 0.05
t / 2,n 1 t0.025,15 2.131