Design and analysis of

experiments
ClassWork 04 – Blocking experiments

Stefania Cacace, PhD Design and Analysis of Experiments 1

 Example from real Lab experience / 2
AMATHO – Additive MAnufacturing of Tiltrotor HOusing

Factors Output
- Batch-2-Batch variability
- Position - Tensile properties
- Orientation (H,45°,V) - Fatigue (Low+High cycle)

#tensile specimens for each condition  5

# fatigue specimens for each condition  20 (ASTM E466)

2 positions x 3 orientations x 5 tensile = 30 specimens

2 positions x 3 orientations x 30 fatigue = 120 specimens

Total = 150 specimens

We need 3 builds

Stefania Cacace, PhD Design and Analysis of Experiments 2

 Example from real Lab experience / 2
No of fatigue No of tensile
Build no Orientation Position
specimens specimens

Vertical
External 6 1 It is not possible to perform a completely
Internal 6 1 randomized experiments, so we need to find a
External 8 3
1 45°
Internal 7 2
way to allocate the specimens in the most
External 7 2 uniform way in the different builds and position
Horizontal
Internal 6 1
External 6 2
Vertical
Internal 8 2
External 8 0
2 45°
Internal 4 2 No design/software can do this !!!!
External 6 2
Horizontal
Internal 8 2
External 8 2
Vertical
Internal 6 2 Manual allocation!
External 4 2
3 45°
Internal 9 1
External 7 1
Horizontal
Internal 6 2
Total 120 30

Stefania Cacace, PhD Design and Analysis of Experiments 3

 Example from real Lab experience / 2

p-values
UTS (MPa) YS (Mpa) Young Mod (GPa) E%
Position 0,600 0,867 0,149 0,172
Orientation 0,000 0,000 0,000 0,008
Build 0,480 0,006 0,318 0,564

Stefania Cacace, PhD Design and Analysis of Experiments 4

 Example from real Lab experience / 2

Stefania Cacace, PhD Design and Analysis of Experiments 5

 Example from real Lab experience / 2

M-ANOVA on the coefficients of the SN curves.

Stefania Cacace, PhD Design and Analysis of Experiments 6

 Theory recap…
Nuisance factors: factor which has an impact on the response, but we are not interested in this
effect.
Controllable factors
• Unknown and uncontrollable:
x1 x2 xp Randomization
…

Output/s • Known but uncontrollable

(Input/s)
Covariance analysis
Process
… • Known and controllable:
Blocking
z1 z2 zp
Uncontrollable factors We design the experiment to control the
(Nuisance factors) variability arising from the nuisance factor.

Stefania Cacace, PhD Design and Analysis of Experiments 7

 Theory recap…

• Randomized complete block design (RCBD): each block is large

enough to hold all the treatments and because the assignment of
each of the levels of the factor in the block is randomized

• Latin square: two controllable nuisance factors

• Randomized incomplete block design (RIBD): when the block is not

large enough to hold all treatmens  any n treatments appear
together an equal number of times

Stefania Cacace, PhD Design and Analysis of Experiments 8

 Exercise 1
An experiment was performed to determine the effect of four different chemicals on the
strength of a fabric. Five fabric samples were selected, and a randomized complete block
design was run by testin geach chemical type once in random order on each fabric sample.

Fabric sample
1 2 3 4 5
1 1,3 1,6 0,5 1,2 1,1
Chemical type

2 2,2 2,4 0,4 2 1,8

3 1,8 1,7 0,6 1,5 1,3
4 3,9 4,4 2,0 4,1 3,4

1) Define the type of design used.

2) Analyze data

Stefania Cacace, PhD Design and Analysis of Experiments 9

 Exercise 1
1. Determine type of exercise
2. Data Snooping: plot data -
3. Perform the analysis with Minitab
4. Verify hypothesis of the model
5. Draw conclusion

Stefania Cacace, PhD Design and Analysis of Experiments 10

 Exercise 1: determine type of exercise
1) Define the type of design used.
It is a randomized complete block design:

Which factor is the block? Fabric

Complete: all combinations of factors have been tested

Randomized: Inside the block, the treatments are tested

in a random order

Stefania Cacace, PhD Design and Analysis of Experiments 11

 Exercise 1: data snooping
2) Analyze data Main Effects Plot for Strength
Data Means
Chemical Fabric sample
Chemical #4 has a strong
3,5
effect on the strength.
3,0
Fabric type #3 shows lower
Mean
2,5 results in respect to the
2,0
other pieces.
1,5
Interaction Plot for Strength
Data Means
1,0
5 Fabric
sample
1 2 3 4 1 2 3 4 5 1
4 2
3
4
5

Blocking desing DO NOT CONSIDER INTERACTION: 3

Mean
we are not interested in the interaction between a 2

blocking factor and the factor of interest.

1
However, draw the interaction plot might be helpful
to verify this assumption. 0
1 2 3 4
Chemical

Stefania Cacace, PhD Design and Analysis of Experiments 12

 Exercise 1: analysis with MTB
The command to perform the analysis is the same used for ANOVA multi-way. Remember NOT to ADD the
interaction between the blocking factor and any other factor.
General Linear Model: Strength versus Fabric sample; Chemical
Analysis of Variance
Source DF Adj SS Adj MS F-Value P-Value
Fabric sample 4 6,6930 1,67325 21,11 0,000
Chemical 3 18,0440 6,01467 75,89 0,000
Error 12 0,9510 0,07925
Total 19 25,6880

Model Summary
S R-sq R-sq(adj) R-sq(pred)
0,281514 96,30% 94,14% 89,72%

Fits and Diagnostics for Unusual Observations

Obs Strength Fit Resid Std Resid
3 0,500 0,055 0,445 2,04 R
18 2,000 2,475 -0,475 -2,18 R
R Large residual

Stefania Cacace, PhD Design and Analysis of Experiments 13

 Exercise 1: residuals assumptions and conclusion
Scatterplot of SRES vs FITS; Fabric sample; Chemical
Probability Plot of SRES
Normal FITS Fabric sample
2
99
Mean -8,63198E-16
StDev 1,026 1
95

90
N
AD
P-Value
20
0,224
0,796
0 We cannot perform test for
80
70
-1 equal variances because
we only have one
Percent

60 -2
50

SRES
40 0 1 2 3 4 1 2 3 4 5
30
20 2
Chemical observation for each
10

5
1 combination of
1
0 conditions.
-3 -2 -1 0 1 2 3 -1
SRES
-2

1 2 3 4

Moreover, the scatterplot revelas dishomogeneity in variances for different fabric samples. However, the
SRES belong to the interval [-2;2]. In conclusion, both chemical type and fabric are significant.
Further experiments are required to verify the influence of the fabric on the response.

Stefania Cacace, PhD Design and Analysis of Experiments 14

 Exercise 2
The effect of 5 different ingredients on the reaction time of a chemical process is being
studied. Each batch of material is only large enough to permit five runs to be made.
Furthermore, each run requires approximately 1 ½ hours, so only five runs can be made in one
day. Analyze data from this experiment using α=5% and draw conclusions.

Day
Batch 1 2 3 4 5
1 A=8 B=7 D=1 C=7 E=3
2 C=11 E=2 A=7 D=3 B=8
3 B=4 A=9 C=10 E=1 D=5
4 D=6 C=8 E=6 B=6 A=10
5 E=4 D=2 B=3 A=8 C=8

a) Determine the type of design c) Find the ingredient which minimizes the reaction time. Use
b) Analyze data Tuckey (evaluate by hand the Tuckey constant and the critical
value, then use Minitab).

Stefania Cacace, PhD Design and Analysis of Experiments 15

 Exercise 2: determine the type of design
a) Determine the type of design

We have two nuisance factors: Batch and day

And one factor of interest: ingredients

Each ingredient has been evaluated in all conditions of day and

batch

It is a latin square design

Stefania Cacace, PhD Design and Analysis of Experiments 16

 Exercise 2: data snooping
b) Analyze data

Main Effects Plot for Reaction time

Data Means
Ingredient seems to be more influecing
Batch Day Ingredient
9 than Batch and Day
8

7
Mean

6
Interaction Plot for Reaction time
Data Means
1 2 3 4 5 1 2 3 4 5
5
10 Batch
1
2
4
3
Batch 5
4
5
3
0
1 2 3 4 5 1 2 3 4 5 1 2 3 4 5 10 Day
1
2
3
Day 5
4

It looks like interactions are significant, however

5

we cannot estimate them.

Ingredient

Stefania Cacace, PhD Design and Analysis of Experiments 17

 Exercise 2: analysis with MTB

General Linear Model: Reaction time versus Batch; Day; Ingredient

Analysis of Variance
Source DF Adj SS Adj MS F-Value P-Value
Batch 4 15,44 3,860 1,23 0,348
Day 4 12,24 3,060 0,98 0,455
Ingredient 4 141,44 35,360 11,31 0,000
Error 12 37,52 3,127
Total 24 206,64

Stefania Cacace, PhD Design and Analysis of Experiments 18

 Exercise 2: residuals assumptions and conclusion
Scatterplot of SRES vs FITS; Batch; Day; Ingredient No outliers (SRES ∈[-2;2]) and no particular patterns
FITS Batch
2
are visible from the scatterplot. The normality cannot
1
be refused (high p-value).
0
Probability Plot of SRES
-1 Normal
99
-2 Mean -1,90958E-16
SRES

2 4 6 8 10 1 2 3 4 5 StDev 1,021
95 N 25
Day Ingredient
AD 0,347
2 90
P-Value 0,452
80
1
70

Percent
60
0
50
40
-1 30
20
-2
1 2 3 4 5 1 2 3 4 5 10
5

1
-3 -2 -1 0 1 2 3
SRES

We conclude that the two nuisance factors do not infuence the response, while
factor ingredient affects the response.

Stefania Cacace, PhD Design and Analysis of Experiments 19

 Exercise 2: comparison
c) Find the ingredient which minimizes the reaction time. Use Tukey (evaluate by hand the Tukey
constant and the critical value, then use Minitab).

1 1 4.508
Tα
= qα ( a,=
df E ) q0.05 ( 5,12
= ) = 3.188
2 2 2
2 MS E 2 ⋅ 3.127
V=Tα 3.188 ⋅
= 3.57
=
n 5

Comparisons for Reaction time

Grouping Information Using the Tukey Method and 95% Confidence
Ingredient N Mean Grouping
3 5 8,8 A
1 5 8,4 A
2 5 5,6 A B
4 5 3,4 B
5 5 3,2 B
Means that do not share a letter are significantly different.

Stefania Cacace, PhD Design and Analysis of Experiments 20

 Exercise 3
An engineer is studying the mileage performance characteristics of five types of gasoline
additives. In the road test he wishes to use cars as blocks; however, because of a time constraint,
he must use an incomplete block design. He runs the balanced design with the five blocks that
follow. Analyze the data from this experiment and draw conclusions α=5%.

Car
Additive 1 2 3 4 5
1 17 14 13 12
2 14 14 13 10
3 12 13 12 9
4 13 11 11 12
5 11 12 10 8

a) Determine the type of design

b) Analyze the data

Stefania Cacace, PhD Design and Analysis of Experiments 21

 Exercise 3: determine the type of design
There is blocking factor: Car
Not all additive are evaluated in each block  Incomplete
The same number of treatment appears in each block  balanced

Car
Additive 1 2 3 4 5
1 17 14 13 12
2 14 14 13 10
3 12 13 12 9
4 13 11 11 12
5 11 12 10 8
4
treatments 4 4 …
(= car)

Stefania Cacace, PhD Design and Analysis of Experiments 22

 Exercise 3: data snooping
Main Effects Plot for Performance
Data Means
Additive Car
14
Both car and additive have
13
an inlfuence on the
response.
Mean

12

Interaction Plot for Performance

11
Data Means

17 Additive
1
10 16 2
3
15 4
1 2 3 4 5 1 2 3 4 5 5
14

13

Mean
12

The interaction looks 11

significant, but cannot be 10

9
estimated. 8

1 2 3 4 5
Car

Stefania Cacace, PhD Design and Analysis of Experiments 23

 Exercise 3: analysis with MTB
General Linear Model: Performance versus Additive; Car
Analysis of Variance
Source DF Adj SS Adj MS F-Value P-Value
Additive 4 35,73 8,9333 9,81 0,001
Car 4 35,23 8,8083 9,67 0,001
Error 11 10,02 0,9106
Total 19 76,95

Stefania Cacace, PhD Design and Analysis of Experiments 24

 Exercise 3: residuals assumptions and conclusion
Probability Plot of SRES
Normal
99

Normality cannot be rejected.

Mean 1,942890E-17
StDev 1,026
95 N 20
AD 0,362
90
P-Value 0,408
80
70

Some dishomogeneity in variance seem to be present at

Percent

60
50
40
30
20 additive #2 and 5. Furher experiments are suggested.
10
5

Scatterplot of SRES vs FITS; Additive; Car

1
-3 -2 -1 0 1 2 3 FITS Additive
2
SRES
1

The ANOVA table shows that both factors are -1

significant. -2

SRES
8 10 12 14 16 1 2 3 4 5

However, the interaction plot looks significant and 2

Car

the homogeoneous variance of SRES should be 1

furher checked.
0

-1

-2

1 2 3 4 5

Stefania Cacace, PhD Design and Analysis of Experiments 25