Six Sigma
IE 4010
Chapter 18:
Design of Experiments (DOE)
Introduction to Designed
Experiments
Often in industry, processes are
“optimized” by varying one variable
at a time.
The problem with this?
− Results take a long time to obtain
− Results are incomplete and/or
confounded
A better method: experimental
design
− All factors are varied together, and
included in the analysis and results
2
Introduction to Designed
Experiments
Experimental design can be used to be more
successful in developing new products or
processes, or improving existing products or
processes
The objective of experimental design is to
purposefully change input variables of a
process, and observe and identify
corresponding changes in the output response
Most experiments involve several variables
− Factorial experimental designs are used in
these situations
3
Application of DOE
A cross functional team can identify potential sources
of variation in the process
During DOE a list of exact settings will be designed
for each trial
− Typically there will be 8 or more runs
− Samples will be randomly taken from each run to
be measured
After DOE samples will be measured, and
calculations performed to determine optimal settings
A confirmation test will determine whether new
settings have improved process as predicted by
results
Note: DOE is best used for
optimizing a process (vs. as a
Basic Terms of DOE
Factor: the variable controlled by the experimenter; it’s influence on a response is studied
in experiment
Levels: The settings of a factor, either quantitative or qualitative
Treatment: A single level assigned to a single factor or experimental unit during the
experimental run (e.g., pressure=200 psi)
Balanced Design: Each setting of each factor appears the same number of times with
each setting of every other factor
Block: A part of the experimental material or environment that is common to itself, and
distinct from other portions
Experimental Design: The formal experiment plan
Experimental Error: Variation in the response when levels and factors are held constant
Planned Grouping: A practice done to promote uniformity within blocks and minimize the
effects of unwanted variables
Randomization: Organizes experiment runs in a “chance” manner
Replication: Repeats observations; requires resetting each trial condition
Repetition: Repeats observations at setup experimental conditions
Dependent Variable: A variable that is affected/changed by another variable
Independent Variable: A variable presumed to affect/change another variable
Response Variable: Show observed results of an experimental treatment, as a result of
levels, interactions, and number of factors
General Guidelines for
Designing Experiments
Recognize (identify) the problem or
opportunity
}
Note: First three
Choose a response variable steps are related to
pre-experimental
Choose factors and their levels planning
Choose the experimental design
Perform the experiment
Analyze the data
Draw conclusions and make recommendations
Note: Checklists on pp 282-283 of text provide
additional guidelines to help ensure successful results! 6
Classification of Experimental
Designs (pp. 317-318)
Completely randomized experiments
Factorials
Blocked factorials
Fractional factorials
Randomized blocks
Latin squares
Response surface
Mixture designs
Factorial Experiments
When there are several factors that
can affect a response variable (y), a
factorial design can be used for an
experiment
− All possible combinations of the
levels of the factors are investigated
− If there are two factors A and B, with
a levels of A, and b levels of B, then
the experiment will contain ab
combinations
− Factorial designs can evaluate main
effects and interactions
8
Types of Factorial Design
2^k design (k factors, each at two
levels)
− k=2 factors
Total number of experimental runs: 4
− k=>3 factors
Total number of experimental runs: 8,
16, 32…
Fractional factorial
− 2^(k-p) designs
− 2^(k-1) designs (half fractional)
For k=2, 2 runs; for k=3, 4 runs…
9
Main Effects
Main effects refer to the change
in response due to changing one
of the primary factors
− They are computed as the
difference between the average
response at the high level of a
factor, and the average response
at the low level
In the figure on this page, the
main effect of factor A is
computed as:
A y A y A (30+40)/2 – (10+20)/2 = 20
10
Main Effects- Another
Example
Notice the subtle difference
in the value on this page:
the value for A+B+ was
changed from 40 to 0
This will change the
computation for the main
effect of factor A as follows:
A y A y A (30+0)/2 - (10+20)/2 = 0
11
Interactions
An interaction between two factors occurs when
the difference in response between the levels of
one factor depends on the setting of another
factor
− If interactions are present, the main effects may
have little meaning
Computing the interaction effect of factor AB for
second example (Fig. 12.6), using experimental
design table:
Run A B AxB Measure
1 - - + 10
2 - + - 20
3 + - - 30
4 + + + 0
AxB y AB y AB (10+0)/2 - (20+30)/2 = -20
12
Graphs of Results
With/Without Interactions
Clearly, we can see an interaction in the case of the
second graph
What do you think would have happened if factors
would’ve been tested one at a time?
13
Design Resolution
The extent to which effects in a fractional factorial design are
aliased with other effects.
− In a fractional factorial design, one or more of the effects
are confounded, meaning they cannot be estimated
separately from one another
Possibilities
− Resolution III
No main effects are aliased with any other main effect, but
main effects are aliased with 2-factor interactions.
− Resolution IV
No main effects are aliased with any other main effect or 2-
factor interactions, but some 2-factor interactions are aliased
with other 2-factor interactions and main effects are aliased
with 3-factor interactions.
− Resolution V
No main effects or 2-factor interactions are aliased with any
other main effect or 2-factor interactions, but 2-factor
interactions are aliased with 3-factor interactions and main
effects are aliased with 4-factor interactions.
14
Statistical Analysis of a Factorial
Experiment using ANOVA
Analysis of Variance is used to
statistically evaluate the results of
an experiment
− For results to be meaningful, trials
should be run in random order
− Care should be taken to minimize
process “noise” external to the
experimental design
15
Data Table for a Two-
Factor Factorial Design
16
Recall: ANOVA Table for a
Two-Factor Factorial Design
17
Calculations for a Two-Factor
Factorial Design
18
DOE Results
Results can be evaluated analytically
− Table of estimated effects and coefficients (t-
and p-values)
− ANOVA
− R^2 value
An equation is constructed using the coefficients for
each factor which represents the relationship
between the response and the factors
Results can also be shown graphically
− Pareto Chart
− Normal plot of standardized results
Residuals can provide information as to data
normality, independence, and whether variances
are equal
19
Steak DOE
You would like to maximize the
approval rating of steak depending
on how it’s prepared and cooked.
The response is the approval rating,
on a scale of 1-100. Your objective is
to maximize this score.
The factors are:
− Cooking method (grill, fry)
− Meat (sirloin, ribeye)
− Marinade (red wine & rosemary, soy
sauce & garlic)
Full-factorial experiment
with replicates
Data from text
Run A B Response (y) Avg
1 - - 28.3 28.6 28.2 28.37
2 - + 33.5 32.7 32.9 33.03
3 + - 24.6 24.6 24.8 24.67
4 + + 37.2 37.6 37 37.27
Analyze these results using Minitab
− Averages only
− All individual values
Popcorn DOE
You would like to optimize your
microwave popcorn, in terms of
minimizing the unpopped kernels in
the bag after it’s cooked.
The three variables you are
interested in looking into are:
− Brand (cheap, costly)
− Time (4 min, 6 min)
− Power (75%, 100%)
DOE In-Class Adventure!
(Modified) Tabletop Hockey
Objective:
learn how to shoot a puck – just like the real gam
e
Response: Furthest distance traveled of a puck
Variables (Factors) and Settings: We decide
− Stick length (short, long)
− Wind up (full, half)
− Shot type (wrist, slap)
− Puck mass (high, low)
Materials: Stick (ruler); Puck: (two types)
Potential Challenges?
23
DOE: Adding Data Points
Replication
− Repeat and replicate measurements are
both multiple response measurements
taken at the same combination of factor
settings
repeat measurements are taken during
the same experimental run
replicate measurements are taken during
identical but distinct experimental runs,
which are often randomized
Center Points
− Experimental runs with all factor levels
set halfway between the low and high
settings.
− Allows you to detect curvature in the
response surface.
24
Response Surface
Methodology
Used to examine the relationship
between a response and a set of
quantitative experimental variables
or factors
Often employed after you have
identified a "vital few" controllable
factors and you want to find the
factor settings that optimize the
response.
Designs of this type can detect
curvature in the response surface.
25