Lean Six Sigma Green Belt
Certification Course
Design of Experiments
Learning Objectives
By the end of this lesson, you will be able to:
Explain the concept of Design of Experiments (DOE)
Compare DOE with regression analysis
Identify the meaning of the basic terms used when
conducting experiments
Explain the effects of DOE error on repetition, replication,
randomization, and blocking
Introduction
The problem
Pre-ground coffee?
Milk?
Added flavors?
Temperature?
Serving size?
JAVA HOUSE COFFEE
Introduction
The solution
Freshly ground
coffee
115 ml milk
5 ml flavoring
60–65⁰C
Serving size 120 ml
JAVA HOUSE COFFEE
Design of Experiments (DOE)
Design of Experiments (DOE)
Input variables Process Output variable
Uncontrollable input
variables
Design of Experiments (DOE)
Is a one-stop method for analyzing all influencing factors simultaneously
Used when known sources of variation have been eliminated but process is
still not capable; if process is capable then DOE may not be required
DOE Identifies which variables individually affect key measures
Determines variable combinations or interactions that impact capability
Provides mathematical model to predict and optimize the response or process
output under recommended settings
Is preferred over One Factor at a Time (OFAT) experiments because it allows
for accelerated learning and is less time-consuming
Regression Analysis
Regression analysis generates a line on a scatter plot that quantifies the relationship between X and
Y to understand variation impact.
High percentage of variability in • Predict future values of Y given X and X given Y
Changes in X
Y (R2> 70%)
• Regress Y on one or more Xs simultaneously
DOE vs. Regression Analysis
DOE Regression
• Determines cause and effect • Uncovers relationships
• Tries bold or creative solutions • Uses existing data
• Clears results on the impact variable • Results left open to interpretation
• Measures impact on variability • Cannot measure variability impact
• Provides reliable information on interactions • Provides risky information on interactions
• Requires leadership support and investment • Does not require leadership support
Basic Terms
Response
Factors
Levels
Interactions
Treatment
Trial
Experimental unit
Basic Terms
Response
Factors
• Dependent variables or outcomes of an experimental treatment that
Levels varies as changes are made to factors
Interactions
Examples:
Treatment
Product strength, average hold time, sales, response time
Trial
Experimental unit
Basic Terms
Response
Factors
• Independent or input variables that are changed during an experiment
Levels to validate their impact on the output
• Can be qualitative and/or quantitative
Interactions
Treatment Examples:
Machine, temperature, procedural change, price
Trial
Experimental unit
Basic Terms
Response Objective: To achieve uniform dimensions for a part at a particular target
value
Factors
Inputs (X) Process Output (Y)
Levels
Interactions Cycle time
Mold temperature
Holding pressure Molding Part hardness
Treatment Holding time
Material type
Trial
Experimental unit Factors Response
Basic Terms
Response
Factors • Settings or conditions of the factors that are tested during the
experiment
Levels • Two levels recommended per factor
• Level 1 is normally coded as “-” and could represent status quo
Interactions • Level 2 is normally coded as “+” and could represent the change
tested
Treatment
Trial Note: The difference between level 1 and level 2 settings should be
significant to detect impact.
Experimental unit
Basic Terms
Response DOE conducted on two factors at two levels to determine the impact
related to plastic part hardness
Factors
Levels Level
Factors
- +
Interactions
Quantitative factor Mold temperature 700° 900°
Treatment
Trial Qualitative factor Plastic type Filler No filler
Experimental unit
Basic Terms
Response
Factors
• When the combination of two factors creates a result that is different
Levels from the result produced by the individual factors
Interactions
Examples:
Treatment Baking time and temperature
Trial
Experimental unit
Basic Terms
Response
Factors
Levels • A unique set of factors at specific levels whose effect on the response
variable is of interest
Interactions
Examples:
Molding temperature at high setting and baking time at low setting
Treatment
Trial
Experimental unit
Basic Terms
Response
Factors
Levels
• An experimental run for a specific treatment
Interactions
Treatment
Trial
Experimental unit
Basic Terms
Response
Factors
Levels
Interactions • Quantity of material to which one trial of a single treatment is applied
to create a response
Treatment
Trial
Experimental unit
DOE Error
Error is variation in experimental units that have been exposed to the same treatment.
Examples:
Humidity | Season | Geographic location | Shift
Repetition and Replication
Repetition Replication
The entire experiment is
During a treatment setup, several
repeated with a change in the
samples are run without
setting of experimental
changing the setting.
conditions between trials.
This shows short-term variability. This shows long-term variability.
Repetition and replication provide an estimate of experimental error and help determine the
statistical significance of the differences in readings.
Error and Randomization and Blocking
• Running the trial without any order
• The use of randomization helps • Setting the DOE so that controllable
with: noise factors are incorporated into the
• Noise factors that are experiment or held at a constant level
completely random and throughout
uncontrollable
• Avoiding time-related changes,
uncontrollable variables, and
tool wear
• Eliminating bias in expert
opinions
Randomization Blocking
Full Factorial Experiment
● Contains all combinations of all levels of all
factors
● Ensures no possible treatment combinations get
Full factorial experimental design
omitted
● Is preferred over other designs
Full Factorial Experiment
Factor B: Factor A:
EXAMPLE Oven Time (X2) Draw Temperature (X1)
700 °F 900 °F
Y1 Y5
30 minutes
Y2 Y6
Y3 Y7
60 minutes
Y4 Y8 The tables depict a
two-way heat
treatment experiment,
Factor A = Temp: Low (-) 700 ⁰F; High (+) 900 ⁰F where Y is the Part
Responses
Factor B = Time: Low (-) 30 minutes; High (+) 60 minutes hardness.
Trial DOE Order A B AB 1 2
1 3 - - + Y1 Y2
2 1 + - - Y3 Y4
3 4 - + - Y5 Y6
4 2 + + + Y7 Y8
Full Factorial Experiment
Main Effect is the average response change going from one setting to the other.
Factor A:
Factor B:
Draw Temperature An analysis of the response helps in understanding how:
Oven Time (X2)
(X1)
700° 900°
A change in A change in oven time Interaction between
temperature at which creates a difference in temperature and time
Y1 Y5 the average part affects the average part
30 minutes the material is drawn
Y2 Y6 hardness hardness
creates a difference in
Y3 Y7 the average part
60 minutes hardness
Y4 Y8
Full Factorial Experiment
Factor B: Factor A:
Oven Time Draw Temperature
A1 = 700° A2 = 900°
B1 = 30 90 84
minutes 87 87
B2 = 60 95 79
minutes 92 78
Results: Results:
The table depicts a two-way heat Changing the draw temperature seems to Changing the oven time seems to
treatment experiment. change the average hardness have no major change in the average
hardness
Full Factorial Experiment
95
Part Hardness
Factor B: Factor A:
Oven Time Draw Temperature
90 700
A1 = 700° A2 = 900°
85
B1 = 30 90 84
minutes 87 87
900
95 79 80
B2 = 60
minutes 92 78
30 min. 60 min.
A1 A2
Results: The interaction plot shows that low
B1 88.5 85.5 temperature and high oven time should be selected
90+87 to achieve the highest desired output of hardness.
A,B, = = 88.5
B2 93.5 78.5 The parallel lines indicate the output if no
2
interactions occur between the main effects.
The mean of the factors
28
Runs
The number of experiments in a DOE setting is known as Runs.
Full Factorial Runs = Levels^factors
The number of runs in a:
Half Fractional Factorial Runs = Levels^(factors-1)
Runs
For Half Fractional factorial experiments, the number of runs is levels to the power of factors minus 1.
Full factorial experiment without replication on 5 factors and 2 levels: 25 = 32
Full factorial experiment with 1 replication on 5 factors and 2 levels: 32 + 32 = 64
Half fractional factorial experiment without replication on 5 factors 25-1 = 16
and 2 levels:
Half fractional factorial experiment with 1 replication on 5 factors 16 + 16 = 32
and 2 levels:
Fractional factorial experiments lose complete information on interactions.
Fun Facts
DID YOU
KNOW…?
The use of DOE is growing in the service
sector.
Key Takeaways
DOE is a structured method that tests input variables and their
impact on the output variable.
Error is variation in experimental units that have been
exposed to the same treatment.
Repetition and replication provide an estimate of
experimental error and help determine the statistical
significance of the differences in readings
Randomization and Blocking can impact the effects of error.
Full factorial experimental design contains all
combinations of all levels of all factors.
The number of experiments in a DOE setting is known as Runs.
Knowledge Check
Knowledge
Check A chef wants to change the recipe for a fondue to improve its taste. He is considering
1 different cheeses and will run 8 experiments. In this scenario, what is the factor?
A. Experiments
B. Fondue
C. Different cheeses
D. Taste
Knowledge
Check A chef wants to change the recipe for a fondue to improve its taste. He is considering
1 different cheeses and will run 8 experiments. In this scenario, what is the factor?
A. Experiments
B. Fondue
C. Different cheeses
D. Taste
The correct answer is C
The factor is the independent variable that is changed and in this case it will be the different cheeses.
Knowledge
Check A chef wants to change the recipe for a fondue to improve its taste. He is considering
2 different cheeses and will run 8 experiments. In this scenario, what is the response?
A. Experiments
B. Fondue
C. Different cheeses
D. Taste
Knowledge
Check A chef wants to change the recipe for a fondue to improve its taste. He is considering
2 different cheeses and will run 8 experiments. In this scenario, what is the response?
A. Experiments
B. Fondue
C. Different cheeses
D. Taste
The correct answer is D
The response is the dependent variable that is impacted by changing factors and in this case it will be the taste.
Knowledge
Check For a full factorial experiment with 3 factors at two levels that are replicated twice,
3 how are trials expected?
A. 24
B. 8
C. 16
D. 9
Knowledge
Check For a full factorial experiment with 3 factors at two levels that are replicated twice,
3 how are trials expected?
A. 24
B. 8
C. 16
D. 9
The correct answer is A
The equation is levels to the power of factors. 2 to the power of 3 is 8 and with two replications would be 8 + 8 + 8 = 24.
