Вы находитесь на странице: 1из 7

Session 13

Recap of Lecture 12

Begin Module 1

Summary

Module 4

Example of a Two Factor Design

Module 3

Design of Experiments with Two Factors

Module 2

Recap of Lecture 12

Module 1

Agenda

Rajiv Gupta
BITS Pilani
Lecture 13

Design of Experiments
Lecture 2

End of Module 1

Lecture 13

The approach takes a PDSA approach to the design of


experiments where at each successive iteration our
degree of belief in the effect of a change is further
increased, or not
The main reason for the departure of the Deming
approach from traditional DOE approaches is that most
studies are analytical rather than enumerative. In
analytical studies, we do not know the behavior of the
system and the population ahead of time. In enumerative
studies the entire population is known and we are
sampling to draw inferences on the entire poupation

Recap of Lecture 12

Lecture 12 was the first lecture on the design of


experiments
In that lecture we discussed that unlike SPC charts,
where we wait for a change to occur in the system, in
design of experiments, we proactively test the effect of
changes to the system
The approach to the design of experiments covered in
lecture 12 did not follow the method in the text book.
However, it is based on the teachings of Deming

Recap of Lecture 12

Interactions among factors


Inefficiency of experimentation
9

One way to do this is to first hold one of the variables


constant, i.e., temperature, and do a single factor
experiment by varying the baking time
Then we could hold the baking time constant and do
a single factor experiment by varying the
temperature
This approach is generally not recommended for two
reasons

Two Factors

Often we might suspect that more than one


variable/factor has an effect on the output or
response variable.
For example, temperature of baking and
length of baking time could affect the way that
cookies turn out, i.e., crispiness, etc.
So how do we test the effect of multiple
factors, starting with 2 factors

Two Factors

Design of Experiments with Two Factors

Begin Module 2

Lecture 13

Factor 1

Factor 2 Value 2

Response

Factor 1

Factor 2 Value 2

Moderate Interaction Between


Factors
Factor 2 Value 1

Response

Factor 2 Value 1

No Interaction Between Factors

As each factor is varied, in turn, the data has to be


collected completely afresh. This takes more time and
effort

Inefficiency of experimentation

12

11

10

Sometimes the effect of varying one of the factors, or both


of them, depends on the value of the other factor that has
been kept constant. When we do one factor at a time
study, these interactions among the factors cannot be
evaluated

Interaction among factors

Two Factors

Factorial Design

Factor 1

Factor 2 Value 1

Factor 2 Value 2

F11, F21
F11, F22
F12, F21
F12, F22

Factorial Design

14

13

15

If the number of factors was 4 and each was to be studied at 2


different levels, then each of the possible factor levels would
have to be combined with each of the other factor levels to
give 2x2x2x2 = 16 experimental units
Quite obviously as the number of factors and/or the number
of levels increases, the number of experimental units also
increases
So for 4 factors, where the first 2 factors are at 2 levels and
the third factor is at 3 levels and the fourth factor is at 5
levels, the number of experimental units = 2x2x3x5 = 60

In a factorial design, the values of both (or more)


variables is varied in combination to permit us to study
interactions among the variables
So for example, if there are 2 variables each at 2 levels,
i.e., F11, F12 and F21 and F22, then we would conduct 4
experiments with the conditions as follows,

Response

Strong Interaction Between Factors

Low
High
Low
High

Low
High
Low
Low
Low
High
High
High

Low
Low
High
Low
High
Low
High
High

Factor
2

Low
Low
Low
High
High
High
Low
High
3 Factors, each at 2 levels

Test
1
2
3
4
5
6
7
8

2 Factors, each at 2 levels

Low
Low
High
High

Factor

Design Matrix

Test
1
2
3
4

Design Matrix

Often it suffices to have factors at two levels, which


we can choose as high and low
In this case, for k factors, the size of the experiment
will be given by 2k
Using 2k designs, we can easily analyze the data
graphically
If we need to have factors at more than 2 levels,
some fractional factorial design may be needed

Factorial Design

18

17

16

Factor 2

Low

High

3 Factors, each at 2 levels

High
Factor 2
Low
High

Low

Factor 1

High

Geometric Display

Factor 1
Low
High

Low
Factor 2
Low
High

Factor 3

Tabular Display

2 Factors, each at 2 levels

Factor 1
Low
High

Factor 2
Low
High

Tabular Display

21

20

19

Low
Low

Factor 1

Lecture 13

Example of a Two Factor Design

Begin Module 3

Lecture 13

End of Module 2

Factor 2

High

High Low

Geometric Display

Factor 3

24

23

22

High

2 levels
2 levels

42% and 48%


175 and 195

Each set of 4 tests will be done with 2 different blends of the


Ingredient. The order of the factor levels is randomized for the 2 blends

Blend 2

Blend 1

Test
1
2
3
4
5
6
7
8

Factor
Temperatur
%
Ingredient
e
42%
195
42%
175
48%
175
48%
195
42%
195
48%
195
42%
175
48%
175

Design Matrix

Replicaions: 2 at each factor level, each with a


different batch of the ingredient

Percentage of ingredient
Temperature (deg C)

Objective: Find a combination of the amount of


ingredient and the reaction temperature to bring th
reaction time to between 7 and 9 hours
Response variable: Reaction time to reach desired
viscosity
Factors:

Example

27

26

25

A chemical manufacturer wants to obtain a particular


viscosity of a product he makes. The viscosity
required that the reaction time be between 7 to 9
hours. Because of a problem, the reaction time for
the process had speeded up with the result that the
viscosity was unacceptable. The manager wanted to
test whether, by varying a key ingredient, and the
reaction temperature, he could slow down the
process to obtain the correct viscosity.

Example

175

9.0, 9.5

9.0, 8.0

2
2
2
2

195
195
175
175

Temperature

5.5, 6.5

Data

42%
48%
42%
48%

6.5
1
9.5
8

Reaction
Time
5.5
9
9
1.5

195

5.5,6.5

1.5,1.0

Blend % Ingredient Temperature


1
42%
195
1
42%
175
1
48%
175
1
48%
195

29

28

30

First, we subtract the reaction time for the low temperature


from the reaction time for the high temperature with the %
ingredient held constant at 48%
1.5 9.0 = -7.5
Next we do the same for % at 42%
5.5 9.0 = -3.5
Holding temperature constant at 175, the difference between
% ingredient at 48 and 42 is
9.0 9.0 = 0
Holding temperature at 195, the difference is
1.5 5.5 = -4.0

Analysis for the First Blend

42%

% Ingredient

48%

5
6
7
8

Test
1
2
3
4

Data

175
9.25

8.5

% Ingredient
42

48

1.25

195

Temperature

Average for the 4 Conditions

33

32

The first thing to note is that the change in the reaction time
to changes in temperature and % of the ingredient were
similar for the two blends. So there is no effect seen of the
two blends
The effect of the temperatures at 48% ingredient
concentration produced a larger difference in reaction time as
compared to that at 42%. This suggests an interaction
between temperature and the concentration of the ingredient
In 3 of the 4 comparisons of the effect of increasing the %
concentration of the ingredient showed an increase in the
reaction time. In one case there was no effect

Analysis

31

First, we subtract the reaction time for the low temperature


from the reaction time for the high temperature with the %
ingredient held constant at 48%
1.0 8.0 = -7.0
Next we do the same for % at 42%
6.5 9.5 = -3.0
Holding temperature constant at 175, the difference between
% ingredient at 48 and 42 is
8.0 9.5 = -1.5
Holding temperature at 195, the difference is
1.0 6.5 = -5.5

Analysis for the Second Blend

42

175

195

48

Conclusion

Concentration

T = 195

T = 175

Analysis

Temperature

48%

42%

35

34

36

From the data and the graphs plotted, it would appear that
the temperature and the concentration of the ingredient have
an effect on the reaction time
Depending on the degree of belief of the people familiar with
the process if the results are conclusive enough to take to the
shop floor, or further testing is required
This is a particular conclusion of Deming with regard to
analytical studies, i.e., the end result is not a purely statistical
one. The conclusion cannot be drawn by a statistician. Only a
person with domain knowledge can interpret the results
accurately.

10

10

Analysis

38

37

39

In this lecture we considered the design of


experiments with more than one factor
One of considerations in multi-factor
experimental design is the number of tests to
be conducted as this involves time and money
The more common way to consider multifactor design of design of experiments is to
consider a 2k design, i.e., each factor at 2
levels

Summary of Lecture 13

Summary

Begin Module 4

End of Module 3

End of Module 4

There are several ways to display the design of a


multi-factor design including the design matrix,
tabular display and the geometric display
We saw an example of a 2 factor design in which
each factor was varied at 2 levels. The results
showed a good correlation between the factors
and the response variable. Also, moderate
interaction between the factors was seen

Summary of Lecture 13

41

40

Вам также может понравиться