Professional Documents
Culture Documents
Reference:
Quality Data Analysis- BM Colosimo
QualityMontgomery –
Data Analysis Chapter 11
1
x1
x2
2
Quality Data Analysis
1
24/05/2021
Multivariate SPC
Control limit computation:
P Allarme 1 In Controllo 1
1
P Allarme 2 In Controllo 2 2
Assume variables to be independent and:
1 2
Generally speaking, for p tests on independent variables:
1 1 p 1 1 1 p
p 2 0.0975 ARL 10
We expected
A few examples: 0.05 p 3 0.143 ARL 7
p 4 0.185 ARL=20
3 ARL 6
Quality Data Analysis
Multivariate SPC
The problem gets more complicated if variables are dependent.
' i
i 1,.., p Bonferroni’s inequality
“familywise”
4
Quality Data Analysis
2
24/05/2021
2 12 ... 1p
Variance-Covariance Matrix 1
22 ... 2p
V ( x ) E [( x E ( x ))( x E ( x ))' ] Σ 12
... ... ... ...
pxp px1 1xp 2p ... 2p
ij E [( x i i )( x j j )] cov( x i , x j )
1p
correlation 1 12 ... 1 p
cov( x i x j ) 1 ... 2 p
P
12
ij
V ( x i )V ( x j ) ... ... ... ...
1 p 2 p ... 1
Correlation matrix
Bivariate nomal distribution:
2
1 x
1
2
f ( x) e - x < +
2 2
2
x
1
x
2
x x - μ ' Σ 1 x - μ
1
1 x - μ ' Σ -1 x - μ
f (x) e 2 - x j < + j = 1,2,..., p
2
p/2
| Σ |1 / 2
6
Quality Data Analysis
3
24/05/2021
X Y
Y
X X
Y Y
Y Y
X X X X
Remind:
X μ
2
X ~ N ( μ, ) 2
Z 2 ~ χ 2 (1)
2
X μ
2
2
X ~ N (μ , ) n Z 2 ~ χ 2 (1)
n 2
Multivariate case:
Has been
X i N p (μ , Σ ) proven that (X μ )'Σ 1 (X μ ) Z 1 2 + Z 2 2 + ...+ Z p 2 ~ 2 ( p )
( i 1, 2 , ..., n )
Has been proven that
X N p ( μ , (1 / n ) Σ ) n (X μ )'Σ 1 (X μ ) Z 1 2 + Z 2 2 + ...+ Z p 2 ~ 2 ( p )
4
24/05/2021
Case p=2
x ' x1 x2 12 12 1 1 2
2
12
Σ Σ
μ ' 1 2 1 2 22 | Σ | 1 2 12
We got:
1 22 1 2 x1 μ 1
n x1 μ 1 x2 μ2 2 2 2 ~ 2 ( p)
1 2 12 2 12 1 x2 μ2
Thus, the statistical quantity:
n
02 2 2 ( x1 1 ) 2 2 1 2 ( x1 1 )( x 2 2 ) 1 2 ( x 2 2 ) 2
1 2 12
2 2 2
A joint control region is applicable:
Alarm if : 02 2 2
2 2
x1 1 x2 2 1 2 x1 1 x 2 2 1 2 2 2 1 2 2 2 ( 2 )
2
1 2 1 2 1 2 1 2 2 2 n
10
Quality Data Analysis
10
5
24/05/2021
1 2
If the ellipse is a circle with radius 2 ( 2 )
n
11
12 0 0 12 0 0
12
Quality Data Analysis
12
6
24/05/2021
x1 Region C
UCL1
Region B
Region A
LCL1
x2
LCL2 UCL2
Region A: out-of-control not signaled by traditional charts
Region B: false out-of-control signaled
Region C: ‘twice’ false out-of-control (because of x1 and x2) signaled
Quality Data Analysis 13
13
2 control chart
A statistical control scheme can be implemented directly into the plane x1 - x2 by
using the control ellipse:
• But the information about the temporal sequence would be lost;
• It would be difficult to depict the control region for 3 variables, and even
impossible for larger numbers of variables
A control chart can be designed to monitor the quantity 0 by using the control
2
scalare
1xp pxp px1
15
7
24/05/2021
T2 control chart
Assume we don’t know and and we m samples of size n to be used in design
phase
Estimators
(considering m samples)
19
In particular:
- Phase 1 (design phase)
T k2 n ( x k x )' S -1 ( x k x ) ~ c 1 ( m , n , p ) F ( p , m ( n 1) ( p 1))
p ( n 1)( m 1)
c1 ( m , n , p ) UCL c 1 ( m , n , p ) F ( p , m ( n 1) ( p 1))
m ( n 1 ) ( p 1)
LCL 0
‐ Phase 2 (future observations):
Under the assumption that m* samples during the design phase
T k2 n ( x k x )' S -1 ( x k x ) ~ c 2 ( m *, n , p ) F ( p , m * ( n 1) ( p 1))
p ( n 1)( m * 1)
c 2 ( m *, n , p ) UCL c 2 ( m *, n , p ) F ( p , m * ( n 1) ( p 1))
m * ( n 1) ( p 1)
LCL 0
20
Quality Data Analysis Quality Engineering
20
8
24/05/2021
c 1 ( m , n , p ) F ( p , m ( n 1) ( p 1)) 2 ( p )
Remind that: 2 ( 1 )
lim F( 1 , 2 )
2 1
21
Example: ultimate
tensile strenght and
diameter of a textile
fiber
n=10; m=20; p=2.
02
n
12 2 2 12 2
2
2
( x1 1 ) 2 2 12 ( x1 1 )( x2 2 ) 1 ( x2 2 ) 2
2
n S 2 ( x x ) 2 2 S ( x x )( x x ) S 2 ( x x ) 2
p 2T2
S1 S 2 S12 2
2 2 2 1 1 12 1 1 2 2 1 2 2
22
Quality Data Analysis
22
9
24/05/2021
No out‐of‐control data.
By using the Phase II limit advocated by Montgomery we have UCL=15.16.
If we used the 2 approximation we would get UCL=13.816, close to the Phase I limit, but
different from the Phase II limit.
23
Problem:
When a control limit violation occurs, how to search for an assignable cause?
A different approach consists of decomposing the T2 statistic into components that reflect the
contribution of each individual variable.
If T2 is the current value of the statistic, and T2(i) is the value of the statistic for all process
variables except the ith one, then:
d i T 2 T (2i ) is an indicator of the relative contribution of the ith variable.
10
24/05/2021
T2 control chart
Example:
Standardized vars:
x ij j
y ij
( m 1) 2j
Out‐of‐control
Cut‐off for di
Quality Data Analysis 25
25
Individual observations
In some SPC applications (e.g., chemical industry) the size of the sample is n=1. In
this case, the Hotelling’s statistic is defined as follows:
The Phase II control limits are:
26
11
24/05/2021
Individual observations
A relevant issue in the presence of individual observations is the variance‐covariance
matrix estimation
The ‘usual’ estimator (long period) is:
This estimator is particularly sensitive to outliers or out‐of‐control data in the original
sample of m observations
Alternative:
A different estimator (short period) is:
27
Quality Data Analysis
27
x 5 . 682 ,88 . 22
12
24/05/2021
Individual observations
Example (Sullivan e Woodall – 1996)
x 5 . 682 ,88 . 22
Example: MEWMA control chart
The ARL value when the process is out‐of‐control increases as p
increases
We are interested in different control‐charting schemes, that are
suitable to monitor processes where the variability is not equally
influenced by all the p variables (the sistem dynamics is driven by
a subset of process variables)
30
Quality Data Analysis
30
13
24/05/2021
Principal Components
Example: (chemical process) 95% confidence region
Scatter plot matrix
32
14
24/05/2021
33
34
Quality Data Analysis
34
15
24/05/2021
Out‐of‐control
35
36
Quality Data Analysis
36
16