QUALITY ENGINEERING

10/09/2018

General recommendations:
 avoid (if not required) theoretical introductions or explanations covered during the course;
 always state the assumptions, formulas/expressions and the final results (when using hypothesis tests provide
the numerical value of the test statistic and the test conclusion in terms of p-value);
 show (qualitatively) all the plots.

Exercise 1 (max score 15)

A grinding process is kept under control by means of an acoustic emission sensor. The average acoustic
emission during each grinding cycle is recorded by means of an integrated monitoring system. The recorded
values in 40 consecutive cycles are reported below (read from left to right and from the top to the bottom).

300.7 305.0 296.7 263.7 210.3 233.0 247.3 265.3 287.0 380.3 341.0 340.0
326.3 342.7 280.0 242.7 273.0 252.0 268.0 278.0 279.7 274.0 338.7 328.7
391.3 405.7 389.7 358.0 356.3 330.3 263.7 273.3 229.7 241.0 244.0 160.3
182.0 203.3 221.3 251.7

a) Design a traditional control chart (assuming a NID behaviour) with run-rules and comment the result.

ONLY FOR MECHANICAL ENGINEERING STUDENTS:

b mech.) Identify and fit a suitable model;
c mech.) Design a control chart based on the fitted model.
d mech.) Estimate the prediction interval for the next grinding cycle.

ONLY FOR MANAGEMENT ENGINEERING STUDENTS:

b man.) Design an EWMA control chart by using the estimator of the standard deviation t based on
the absolute value of the prevision errors, with .
c man.) Verify that the EWMA is a good one-step-ahead predictor for the acoustic emission data.

Exercise 2 (max score 11)

An electron beam welding process is monitored by means of an in-situ sensor that acquires the X-ray emission
during the process. Three signal descriptors are extracted from the raw signal during each welding operation.
In order to design a control chart, 25 consecutive welds were monitored: the resulting data are shown in the
table below:

X Y Z
1 0,94 13,92 22,47
2 1,31 27,83 28,33
3 1,06 13,94 25,78
4 1,16 12,20 25,18
5 1,32 12,95 25,85
6 1,40 2,48 21,81
7 1,00 9,62 24,92
8 1,44 28,67 29,47
9 1,34 13,18 26,42
10 1,21 8,47 25,49
11 1,19 0,09 23,4
12 1,40 14,95 25,89
 13 1,17 3,91 26,35
14 1,23 36,55 26,41
15 1,23 28,02 24,17
16 1,37 7,91 25,05
17 1,48 5,65 30,56
18 1,31 7,65 20,52
19 1,19 0,59 30,49
20 1,33 2,74 26,6
21 1,26 1,84 27,99
22 1,22 4,42 20,96
23 1,39 9,69 24,11
24 1,28 15,98 24,71
25 1,25 3,63 26,35

A set of 5 new observations was then acquired:

X Y Z
26 1,32 33,79 21,7
27 1,27 22,78 26,67
28 1,04 5,53 21,44
29 1,19 12,25 25,46
30 1,2 2,53 27,37

ONLY FOR MECHANICAL ENGINEERING STUDENTS

a mech.) Design a control chart for the mean based on the long-term variance-covariance estimator and
discuss the result (use only the first 25 observations).
b mech.) By using the control chart designed at the previous point, determine if the 5 new observations
are in-control or not.
c mech.) Is the mean of the 5 new observations of variable X statistically different from the one
observed in Phase I? (use an appropriate statistical test).

ONLY FOR MANAGEMENT ENGINEERING STUDENTS

a man.) Propose a method to reduce the dimensionality of the problem in such a way to capture at least
75% of the overall variability. How many principal components are needed? Discuss the
results (show eigenvalues and eigenvector values).
b man.) Design an appropriate control chart based on the result of the previous point (use only the first
25 observations).
c man.) By using the control chart designed at the previous point, determine if the 5 new observations
are in-control or not.

Exercise 3 (max score 4)

An X - S control chart is used to monitor a dimensional characteristic in a discrete manufacturing process
(sample size n=5, Type I error α=0.005). The characteristic is known to be normally distributed with a mean
value of 0.75 and a standard deviation of 0.12 s.

Assuming that an out-of-control shift of the mean is detected with a Type II error equal to 0.865, compute
 the entity of the shift in terms of standard deviation units
 the out-of-control mean.
Solution
Exercise 1
a)
Time-series plot:

Time Series Plot of X

400

350

300
X

250

200

150
4 8 12 16 20 24 28 32 36 40
Index

The process seems not NID, but let’s design the traditional chart assuming a NID behaviour:
I-MR Chart of X
1
400 1
1 1

5 5
5 5 5 UCL=358,7
6 6
Individual Value

300 _
X=286,4

2
5
2 LCL=214,1
200 1
1
1
1
1 5 9 13 17 21 25 29 33 37
Observation

1
UCL=88,8
80
Moving Range

60

40
__
MR=27,2
20

0 3 LCL=0
1 5 9 13 17 21 25 29 33 37
Observation

Run rules:
Test Results for I Chart of X

TEST 1. One point more than 3,00 standard deviations from center line.
Test Failed at points: 5; 10; 25; 26; 27; 36; 37; 38

TEST 2. 9 points in a row on same side of center line.

Test Failed at points: 39; 40

TEST 5. 2 out of 3 points more than 2 standard deviations from center

line (on one side of
CL).
Test Failed at points: 6; 11; 12; 14; 25; 26; 27; 28; 29; 37; 38; 39
TEST 6. 4 out of 5 points more than 1 standard deviation from center
line (on one side of
CL).
Test Failed at points: 13; 14; 26; 27; 28; 29; 30; 36; 37; 38; 39; 40

TEST 8. 8 points in a row more than 1 standard deviation from center

line (above and below
CL).
Test Failed at points: 30; 40

Test Results for MR Chart of X

TEST 1. One point more than 3,00 standard deviations from center line.
Test Failed at points: 10

TEST 3. 6 points in a row all increasing or all decreasing.

Test Failed at points: 21

ONLY FOR MECHANICAL ENGINEERING STUDENTS

b mech)
Runs-test:
Runs test for X

Runs above and below K = 286,392

The observed number of runs = 6

The expected number of runs = 20,55
17 observations above K; 23 below
P-value = 0,000

ACF and PACF:

Autocorrelation Function for X

(with 5% significance limits for the autocorrelations)

1,0

0,8

0,6

0,4
Autocorrelation

0,2

0,0

-0,2

-0,4

-0,6

-0,8

-1,0

1 2 3 4 5 6 7 8 9 10
Lag
 Partial Autocorrelation Function for X
(with 5% significance limits for the partial autocorrelations)

1,0

0,8

0,6

Partial Autocorrelation
0,4

0,2

0,0

-0,2

-0,4

-0,6

-0,8

-1,0

1 2 3 4 5 6 7 8 9 10
Lag

The process is not random, and a suitable model may be AR(1).

Analysis of Variance

Source DF Adj SS Adj MS F-Value P-Value

Regression 1 89221 89221 73,40 0,000
AR1 1 89221 89221 73,40 0,000
Error 37 44976 1216
Lack-of-Fit 36 42991 1194 0,60 0,794
Pure Error 1 1984 1984
Total 38 134196

Model Summary

S R-sq R-sq(adj) R-sq(pred)

34,8648 66,49% 65,58% 63,84%

Coefficients

Term Coef SE Coef T-Value P-Value VIF

Constant 50,9 28,0 1,82 0,077
AR1 0,8185 0,0955 8,57 0,000 1,00

Regression Equation

X = 50,9 + 0,8185 AR1

The constant term is not significant. Let’s remove it.

Analysis of Variance

Source DF Adj SS Adj MS F-Value P-Value

Regression 1 3275826 3275826 2541,14 0,000
AR1 1 3275826 3275826 2541,14 0,000
Error 38 48986 1289
Lack-of-Fit 37 47002 1270 0,64 0,781
Pure Error 1 1984 1984
Total 39 3324812

Model Summary

S R-sq R-sq(adj) R-sq(pred)

35,9043 98,53% 98,49% 98,45%

Coefficients

Term Coef SE Coef T-Value P-Value VIF

AR1 0,9886 0,0196 50,41 0,000 1,00

Regression Equation

X = 0,9886 AR1

The model without constant meet the assumptions (normality of residuals: p-value=0.118, runs-test: 0.294,
ACF & PCAF ok, no strange pattern, lack of fit ok).
 Autocorrelation Function for RESI_1 Partial Autocorrelation Function for RESI_1
(with 5% significance limits for the autocorrelations) (with 5% significance limits for the partial autocorrelations)

1,0 1,0

0,8 0,8

0,6 0,6

Partial Autocorrelation
0,4 0,4
Autocorrelation

0,2 0,2

0,0 0,0

-0,2 -0,2

-0,4 -0,4

-0,6 -0,6

-0,8 -0,8

-1,0 -1,0

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Lag Lag

The model is approximately a random walk.

c mech)
Special cause control chart:
I-MR Chart of RESI_1
UCL=108,2
100

50
Individual Value

_
0 X=2,0

-50

-100 LCL=-104,1
1 5 9 13 17 21 25 29 33 37
Observation

150
1
UCL=130,4

100
Moving Range

50 __
MR=39,9

0 LCL=0
1 5 9 13 17 21 25 29 33 37
Observation

One out-of-control point is signalled by the MR chart only. It can be caused by the fact that the MR statistic
follows an half-normal distribution. Let’s transform it to normality with the Box-Cox transformation:
The result yields a value close to the known transformation (λ=0.4). By using λ=0.4, the new MR chart is:

I Chart of C9
8
UCL=7,730

6
Individual Value

_
4 X=3,946

LCL=0,161
0
1 5 9 13 17 21 25 29 33 37
Observation

The process is in control.

d mech)
95% Prediction interval:

Variable Setting
AR1 251,7

Prediction
Fit SE Fit 95% CI 95% PI
248,828 4,93612 (238,836; 258,821) (175,460; 322,196)
ONLY FOR MANAGEMENT ENGINEERING STUDENTS
b man)

The EWMA control chart for auto-correlated data to be used is the following:

Xˆ t 1|t  zt  xt  (1   ) zt 1 ( z0  x )

m | et |
 t   | et | (1   ) t 1   0.1  ( 0)   ̂ t  1.25  t
t 1 m

The λ parameter can be estimated by minimizing the SSE: the result is λ=0.975.
The resulting EWMA control chart is:

There is one out-of-control observation at time t=10. We have no information about the existence of
assignable causes, thus the EWMA control chart design is over. That OOC observation deserves some
attention.

c man)

In order to determine if the EWMA is a good one-step-ahead predictor we should check its residuals:
Normality can be accepted (at alpha=5%).
Runs test:
Runs test for RES-EWMA

Runs above and below K = -0,910030

The observed number of runs = 19

The expected number of runs = 20,8
22 observations above K; 18 below
P-value = 0,560

ACF and PACF:

Autocorrelation Function for RES-EWMA Partial Autocorrelation Function for RES-EWMA

(with 5% significance limits for the autocorrelations) (with 5% significance limits for the partial autocorrelations)

1,0 1,0

0,8 0,8

0,6 0,6
Partial Autocorrelation

0,4 0,4
Autocorrelation

0,2 0,2

0,0 0,0

-0,2 -0,2

-0,4 -0,4

-0,6 -0,6

-0,8 -0,8

-1,0 -1,0

1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Lag Lag

The residuals are ok. The EWMA is a suitable model.

EXERCISE 2
Time-series plots and scatter plots:

Time Series Plot of X; Y; Z

1 5 10 15 20 25
X Y
40
1,4
30

1,2 20

10
1,0
0
Z
30,0

27,5

25,0

22,5

20,0
1 5 10 15 20 25
Index
 Scatterplot of X vs Y; X vs Z; Y vs Z
X*Y X*Z

1,4

1,2

1,0

0 10 20 30 40
20,0 22,5 25,0 27,5 30,0

Y*Z
40

30

20

10

0
20,0 22,5 25,0 27,5 30,0

Let’s check the assumptions.

Randomness:

Runs Test: X; Y; Z
Descriptive Statistics
Number of
Observations
Variable N K ≤K >K
X 25 1,2587 13 12
Y 25 11,4759 14 11
Z 25 25,5712 12 13
K = sample mean
Test
Null hypothesis H₀: The order of the data is random
Alternative hypothesis H₁: The order of the data is not random
Number of Runs
Variable Observed Expected P-Value
X 15 13,48 0,534
Y 10 13,32 0,168
Z 14 13,48 0,831
 Autocorrelation Function for X Partial Autocorrelation Function for X
(with 5% significance limits for the autocorrelations) (with 5% significance limits for the partial autocorrelations)

1,0 1,0

0,8 0,8

0,6 0,6

Partial Autocorrelation
0,4 0,4
Autocorrelation

0,2 0,2

0,0 0,0

-0,2 -0,2

-0,4 -0,4

-0,6 -0,6

-0,8 -0,8

-1,0 -1,0

1 2 3 4 5 6 1 2 3 4 5 6
Lag Lag

Autocorrelation Function for Y Partial Autocorrelation Function for Y

(with 5% significance limits for the autocorrelations) (with 5% significance limits for the partial autocorrelations)

1,0 1,0

0,8 0,8

0,6 0,6

Partial Autocorrelation
0,4 0,4
Autocorrelation

0,2 0,2

0,0 0,0

-0,2 -0,2

-0,4 -0,4

-0,6 -0,6

-0,8 -0,8

-1,0 -1,0

1 2 3 4 5 6 1 2 3 4 5 6

Lag Lag

Autocorrelation Function for Z Partial Autocorrelation Function for Z

(with 5% significance limits for the autocorrelations) (with 5% significance limits for the partial autocorrelations)

1,0 1,0

0,8 0,8
0,6 0,6
Partial Autocorrelation

0,4 0,4
Autocorrelation

0,2 0,2

0,0 0,0

-0,2 -0,2

-0,4 -0,4

-0,6 -0,6

-0,8 -0,8

-1,0 -1,0

1 2 3 4 5 6 1 2 3 4 5 6

Lag Lag

There is no statistical evidence of violations of the randomness assumption.

Normality (marginal):
The signal descriptor Y violates the marginal normality assumption. Let’s try to apply Box-Cox:

If we set 𝜆 = 0.5, normality is acceptable at 5%:

ONLY FOR MECHANICAL ENGINEERING STUDENTS:
a mech)

The T2 control chart for the Phase I data after the transformation of Y is:

T² Chart of X; ...; Z
16
UCL=15,51

14

12

10

8
T²

4 Median=3,92

1 3 5 7 9 11 13 15 17 19 21 23 25
Sample
b mech)
Before applying the control chart to the new observation, the descriptor Y must be transformed with
the same exponential used in Phase I (𝜆 = 0.5). The dataset after the transformation is:

X Y Z
26 1,32 5,81262 21,7
27 1,27 4,77236 26,67
28 1,04 2,3515 21,44
29 1,19 3,500465 25,46
30 1,20 1,589665 27,37

The resulting control chart is:

T² Chart of Xall; ...; Zall

16
UCL=15,51

14

12

10

8
T²

4 Median=3,92

1 4 7 10 13 16 19 22 25 28
Sample
Subgroups omitted from the calculations: 26-30

The new observation are in-control.

c mech)

A statistical test can be performed after checking the normality and independence of the new
observations of X. However, due to the very small sample size, we can simply assume that it was
drawn from a normal and independent population.
Then, a test to determine is the two variances are statistically equal or not is needed:

Descriptive Statistics
Variable N StDev Variance 95% CI for σ
 X 25 0,130 0,017 (0,096; 0,191)
Xnew 5 0,105 0,011 (0,044; 0,413)

Test
Null hypothesis H₀: σ₁ / σ₂ = 1
Alternative hypothesis H₁: σ₁ / σ₂ ≠ 1
Significance level α = 0,05
Test
Method Statistic DF1 DF2 P-Value
Bonett * 0,697
Levene 0,58 1 28 0,454

Since there is no statistical difference in terms of variance, a 2-sample t test for the means with
equal variances can be performed. The results is the following:

Two-Sample T-Test and CI: X; Xnew

Method
μ₁: mean of X
µ₂: mean of Xnew
Difference: μ₁ - µ₂
Equal variances are assumed for this analysis.
Descriptive Statistics
Sample N Mean StDev SE Mean
X 25 1,259 0,130 0,026
Xnew 5 1,204 0,105 0,047
Estimation for Difference
Pooled 95% CI for
Difference StDev Difference
0,0550 0,1272 (-0,0726; 0,1827)
Test
Null hypothesis H₀: μ₁ - µ₂ = 0
Alternative hypothesis H₁: μ₁ - µ₂ ≠ 0
T-Value DF P-Value
0,88 28 0,385

There is no statistical difference between the means of the two samples.

ONLY FOR MANAGEMENT ENGINEERING STUDENTS:

a man)

Let’s apply the PCA based on Correlation Matrix, as the three descriptors are defined on different
scales.

Principal Component Analysis: X; Ytrans; Z

Eigenanalysis of the Correlation Matrix
Eigenvalue 1,2895 0,9748 0,7357
Proportion 0,430 0,325 0,245
Cumulative 0,430 0,755 1,000

Eigenvectors
Variable PC1 PC2 PC3
X 0,662 0,302 -0,686
Ytrans 0,303 -0,945 -0,123
Z 0,686 0,126 0,717

Scatterplot of PC1; PC2; PC3 vs C16

PC1*C16 PC2*C16
0,7

0,6 0,0

0,5
-0,5
0,4

0,3 -1,0
X Y Z X Y Z

PC3*C16
0,8

0,4

0,0

-0,4

-0,8
X Y Z

In order to retain at least 75% of the overall variability, the first two PCs are retained.
b man)

T2 chart with long term variance-covariance estimator:

T² Chart of Score 1; Score 2

14
UCL=13,62

12

10

8
T²

2 Median=2,29

1 3 5 7 9 11 13 15 17 19 21 23 25
Sample
Subgroups omitted from the calculations: 26-30

The process is in-control.

c man)
The new 5 observations must be projected onto the first two principal components, but, first, the
new observations of variable Y must be transformed with the same exponential used in Phase I (𝜆 =
0.5). The dataset after the transformation is:

X Y Z
26 1,32 5,81262 21,7
27 1,27 4,77236 26,67
28 1,04 2,3515 21,44
29 1,19 3,500465 25,46
30 1,20 1,589665 27,37

Since, the correlation matrix was used, the scores are computed by standardizing the original data.
The Phase I means and standard deviations are the following:
 𝑋̅ = 1,26 𝑠𝑥 = 0,13
𝑌̅ = 3,07 𝑠𝑦 = 1,46
𝑍̅ = 25,57 𝑠𝑧 = 2,60

By standardizing the new 5 observations with respect to the above means and standard deviations and
projecting them along the first two PCs, the resulting T2 control chart is the following:

T² Chart of Score1all; Score2all

14
UCL=13,62

12

10

8
T²

2 Median=2,29

1 4 7 10 13 16 19 22 25 28
Sample
Subgroups omitted from the calculations: 26-30

The new observations are in-control.

Exercise 3
Xbar-S control chart with known parameters:

LCS    k / n LCS   S  k S  c4  k 1  c4 2 

LC   LC   S  c4
LCI    k / n LCI   S  k S  c4  k 1  c4 2 

Alpha 0,005
Alpha/2 0,0025
K=z_alpha/2 2,807
n 5
c4(5) 0,94
 X-bar chart S-chart
UCL CL LCL UCL CL LCL
0,901 0,75 0,599 0,228 0,113 0

Let’s compute the operating characteristic curve:

H1: 𝜇𝑛𝑒𝑤 = 𝜇 + 𝛿𝜎

Xbar chart:

𝑈𝐶𝐿𝑋𝑏𝑎𝑟 − 𝜇𝑛𝑒𝑤 𝐿𝐶𝐿𝑋𝑏𝑎𝑟 − 𝜇𝑛𝑒𝑤

𝛽𝑋̅ = 𝑃(𝐿𝐶𝐿𝑋𝑏𝑎𝑟 ≤ 𝑋̅ ≤ 𝑈𝐶𝐿𝑋𝑏𝑎𝑟 |𝐻1 ) = 𝑃 (𝑍 ≤ ) − 𝑃 (𝑍 ≤ )
𝜎/√𝑛 𝜎/√𝑛

This is a function of delta.

S chart:

(𝑛 − 1) (𝑛 − 1) 2 (𝑛 − 1)
𝛽𝑆 = 𝑃(𝐿𝐶𝐿𝑆 ≤ 𝑆 ≤ 𝑈𝐶𝐿𝑆 |𝐻1 ) = 𝑃 ( 2
𝐿𝐶𝐿𝑆 2 ≤ 𝑆 ≤ 𝑈𝐶𝐿𝑆 2 )
𝜎 𝜎2 𝜎2

This is constant.

Eventually:
𝛽 = 𝑃(𝑛𝑜 𝑎𝑙𝑎𝑟𝑚|𝐻1 ) = 𝑃(𝑛𝑜 𝑎𝑙𝑎𝑟𝑚 𝑓𝑟𝑜𝑚 𝑋𝑏𝑎𝑟 𝑐ℎ𝑎𝑟𝑡|𝐻1) ∗ 𝑃(𝑛𝑜 𝑎𝑙𝑎𝑟𝑚 𝑓𝑟𝑜𝑚 𝑆 𝑐ℎ𝑎𝑟𝑡|𝐻1 )

By computing the Type II error for different values of 𝛿, we can see that 𝛽 = 0,865 when 𝛿 = 0,8
standard deviation units. Thus, the out-of-control mean is 𝜇𝑛𝑒𝑤 = 0,846.