Design of Experiments
Chapter 1 - Introduction
Basic concepts of DOE
– Definition.
– Process.
– Applications of DOE.
– Terminology.
– Principles.
– Comparison of Classical and Statistically Designed
Experimentation.
– Guidelines for Designing Experiments.
– Some Important Experimental Designs.
Definitions of DOE
• Douglas C Montgomery
– A designed experiment is a test in which purposeful
changes are made to the input variables of a process
or a system to identify the reasons for changes in the
output response.
• Taguchi
– Design of experiments is a systematic and efficient
way of designing high quality products and processes
at low cost.
Process
Applications of DOE
• Characterizing a process.
• Optimizing a process.
• Product Design.
• Determining System and Component tolerances.
DOE Terminology
• Independent Variables-X’s
– Also called factors
– Factors or variables we select in advance
– The causes
• Dependent Variables-Y
– Also called responses
– The quantity (Y) that we measure to determine the impact of the X’s
– The effect.
DOE Terminology
For Example: Capital Logistics GB
• Levels Y=on-time delivery of shipments
– The test settings for X X1=Truck size
Levels=62ft & 66ft truck.
X2= The Type
Levels=Brand X & Brand Y
• Main effects For Example: Is the on-time delivery different
for:
– Differences between each a. 62ft trucks vs 66ft trucks?
factor level b. Brand X nor Brand Y tires?
For Example: Is on-time delivery different
• Interactions when:
a. 62ft truck, Brand X tire
– Differences between two or b. 62ft truck, Brand Y tire.
more factor level c. 66ft truck, Brand X tire
combinations d. 66ft truck, Brand Y tire.
X
Truck
Truck
66 ft
62 ft
Y
Interactions
Tire Type
DOE Terminology
Deviation
from
required
Delivery time
Y
X
X
Y
Tire Type
Truck Size
66
Main Effects
X
62
Deviation
Deviation from
from required
required Delivery time
Delivery time Y
Y
Basic Principles of DOE
• Replication.
• Randomization.
• Error Control or Blocking.
Comparison of Classical and Statistically
Designed Experimentation
Criterion Classical Experimentation Statistically Designed
Experimentation
Principle Common Sense Statistical Principles
Factor One at a time Intelligent Multifactor
Manipulation Manipulation
Cost Effectiveness Expensive Cost Effective
Skill Required Not High Highly Skillful
Applicability For Laboratory Experiments Laboratory and Real Life
only Situations
Error Concept No error Concept Error is Quantified
Interaction Not Imagined Calculated
Validity of Not Tested for Significance Tested for Signifiance
Conclusions
Extrapolation Required Not Required
Guidelines for Designing Experiments
1. Recognition and Statement of the Problem.
2. Choice of Factors and Levels.
3. Selection of the Response Variable.
4. Choice of Experimental Design.
5. Performing the Experiment.
6. Data Analysis.
7. Conclusions and Recommendations.
Some Important Experimental Designs
• Completely Randomized Designs.
• Randomized Block Designs.
• Latin Square Designs.
• Factorial Designs.
• Fractional Factorial Designs.
• Orthogonal Array Designs.
• Robust Design.
Introduction to Basic Designs
Completely Randomized Design
Best suited when:
• Experimental units are homogenous.
• The Pattern of homogeneity not known.
Advantages:
• Any number of factor classes and replication may be used.
• The statistical analysis is relatively simple.
Disadvantages:
• Experimental units variability cannot be isolated form error.
• Allows one way analysis of variance and enables only one
conclusion.
Example
Tensile strength lb/in2
% of 1 2 3 4 5
cotton
15 7 7 15 11 9
20 12 17 12 18 18
25 14 18 18 19 19
30 19 25 22 19 23
35 7 10 11 15 11
ANOVA
Source Degrees Sum of Mean FCAL FTAB
of squares squares
F(0.05,4,20)
freedom
Due to
treatments 4 475.76 118.94 14.76 2.87
(% of
cotton)
Due to 20 161.20 8.06 ----- -----
error
Total 24 693.96 ----- ---- -----
Data Analysis
Total =376
N =no. of observations
Correction factor (CF)=T2/N
=(376)2/25
=5655.04
SSTreatments =ΣCj2/5-CF
=(492 +772+882+1082+542)/5 - 5655.04
=475.76
SSTotal = Σ Σ X2 – C.F
=636.96
Data Analysis
SS Error = SSTotal – SSTreatments
= 636.96 – 475.76
=161.20
MS Treatments =SSTreatments/df
=475.76/4 = 118.94
MS Error =SSError/df
=161.20/20 =8.06
Fcal = MS Treatments/MS Error
=118.94/8.06 = 14.76
Ftab = F (0.05,4,20)
TESTING OF HYPOTHESIS
HO : % Cotton content does not affect tensile strength
of the synthetic fiber.
i.e. μ1=μ2=μ3=μ4=μ5
HI : % of cotton content affect tensile strength of the
synthetic fiber.
i.e. μ1≠μ2≠μ3≠μ4≠μ5
Since FCAL > FTAB HO is rejected and may be concluded that
averages differ i.e. Percentage of cotton in the synthetic fiber
significantly affects the average tensile strength.
Box plot of tensile strength v/s cotton
percentage
Point /line chart for tensile strength v/s cotton
percentage
Introduction to Basic Designs
–Randomized Block Design
• Useful to control variability arising from known nuisance sources
(raw material, people, time).
• It is two way analysis of variance.
• Used in investigating into two factors simultaneously or one factor
with a nuisance variable.
• Separates variability between specimens from the experimental
error.
• Heterogeneity is controlled by applying the treatments over blocks
of homogenous material.
• Allows any number of treatments and any number of replicates.
Example
Type of tip
Coupons 1 2 3 4
(blocks)
1 9.3 9.4 9.2 9.7
2 9.4 9.3 9.4 9.6
3 9.6 9.8 9.5 10.0
4 10.0 9.9 9.7 10.2
ANOVA
Sources Degrees Sum of Mean FCAL FTAB
of of squares squares
variation freedom
Due to 3 0.385 0.128 14.38 3.86
treatme
nts
Due to 3 0.825 0.275 30.89 3.86
blocks
Due to 9 0.08 0.009 ----- -----
error
Total 15 1.29 ----- ----- -----
Data Analysis
2
CF = T /N
= 1542/16
=1482.25
SS TOTAL
=
=1483.54 – 1482.25
=1.29
SS TIPS =
=1482.635 – 1482.25
=0.385
Data Analysis
SS COUPONS =
= 1483.075 - 1482.24
=0.825
F CAL(TIPS) = MS treatments/MS error
= 0.128/0.0089
= 14.38
F CAL(coupons) = MS coupons/MS error
= 0.275/0.0089
= 30.89
F TAB =F(0.05,3,9)
= 3.86
TESTING OF HYPOTHESIS :
HO = μ1=μ2=μ3=μ4
HI = μ1≠μ2≠μ3≠μ4
Since F CAL > FTAB HO is rejected. It may be concluded that type of
tip affects the mean hardness reading. Also the blocks
(coupons) differ significantly since the mean square for blocks
is larger relative to error.
Point /line chart for type of tip v/s hardness
Point /line chart for coupon v/s hardness
Interaction chart of hardness v/s coupons
