Notre Dame University – Louaize
Faculty of Engineering
Department of Civil and Environmental Engineering Fall 2012
CEN 431: Concrete & Pavement Design LAB
Design of an Engineering Experiment
I. What is an Experiment?
An experiment is a research method utilized for testing different assumptions or
hypothesis, through trial and error under conditions that are constructed and controlled by the
researcher. During the experiment, one or more conditions, which are referred to as the
independent variables, are allowed to vary in an organized manner and the effects of these
variations on associated conditions, which are referred to as the dependant variables, are
measured, recorded, validated and analyzed in order to reach a conclusion.
II. Utilization of an Experiment:
When a process needs to be analyzed, experiments are often used to evaluate which
process inputs have significant impact on the process output, and what the target levels of
those inputs should be to attain a desired output.
III. Role of Designing an Experiment:
1. Conduct a relatively small number of tests in order to isolate the most important
factors.
2. Determine whether any of the factors interact, and if they do, find the level of
interaction since combined effects are as important as individual ones.
3. Anticipate the response for any combination of factors.
IV. Definitions:
1. Factor: A controllable experimental variable thought to influence the response.
2. Response: The outcome or result being interested in determining.
3. Levels: Specific values for the factor which can be the lower, higher, or any value in
between.
4. Interaction: Factors may have independent effects on the response of an experiment,
while some may interact between each other to produce an effect. (Those are hard to
identify and are of least importance).
5. Replication: Deals with gathering of results in a repeated manner for a set of levels
for all the factors. Repeatability concerns are answered by such experiments.
V. Steps of Designing an Experiment:
1. Define: to define goals and objectives of the experiment.
2. Research: to investigate how the factors chosen will affect the outcome through
internet or published researches…
3. Select the variables: in an experiment, there are two types of variables, independent
variable and dependant variable. The independent variable is the one that is varied
or manipulated by the researcher, whereas the dependant variable is the response that
CEN 431 Design of an Engineering Experiment 1/4
is measured. In other words, an independent variable is the presumed cause, while the
dependant variable is the presumed effect.
4. Select appropriate methods for measuring variables.
5. Select proper range of independent variables.
6. Determine appropriate number of data points (Levels).
VI. Types of Design
There are three types of designs for an experiment:
1. Simple Design
2. Full Factorial Design
3. Fractional Factorial Design
A. Simple Design
• Vary one factor at a time while fixing all other factors
• Assume that none of the factors interact with each other
• A large number of experiments is required
B. Full Factorial Design
• Test every combination of factor levels
• A large number of experiments are needed (Disadvantage)
• All interactions are captured (Advantage)
Using the simplest example, a 2-factors/2-levels example can help in generalizing an n-
factors/n-levels example.
Example: Measuring the compressive strength of concrete (𝑓𝑓𝑐𝑐′ )
The two factors that are going to be considered are: Factor A: Curing Period (xA) and Factor B:
Water to Cement Ratio (xB):
High Value: +1
Factor A (xA): Curing period
Low Value: -1
High Value: +1
Factor B (xB): w/c Ratio
Low Value: -1
Usually high values are tested for as +1 and low values as -1, since it is more convenient to work with
such numbers yielding a mean equal to zero.
y is the outcome of the experiment (compressive strength):
Therefore: y = f (xA, xB, xAB)
CEN 431 Design of an Engineering Experiment 2/4
Where: xAB is the effect of both factors A and B (interaction)
Converting to matrix:
xA -1 +1
xB -1 +1
The number of experiments is computed according to the following formula: n = (levels)facto
rs
.
In this case n = 22 = 4 experiments.
Exp #1both factors have low levels: -1 -1
Exp#2 factor A is low, B is high: -1 +1
Exp#3 factor A is high, B is low: +1 -1
Exp#4 both factors have high levels: +1 +1
Converting the experiments into a matrix of factors and levels combinations we get:
xA xB xAB(product)
-1 +1 -1
+1 -1 -1
-1 -1 +1
+1 +1 +1
Sum= 0 0 0
Assume a linear relationship:
y = P0 + PAxA + PBxB + PABxAB
For each experiment one outcome is obtained, the compressive strength of concrete, therefore there
must be 4 equations and 4 unknowns:
xA - +
xB
- 3000 psi 1500 psi (1) 3000 = P0 - PA - PB + PAB
(2) 3250 = P0 - PA + PB - PAB
(3) 1500 = P0 +PA - PB - PAB
+ 3250 psi 4000 psi (4) 4000 = P0 + PA + PB + PAB
By solving the 4-equation 4-unknown problem in question, the following values can be obtained:
P0 = 2937.5
PA = -187.5
PB = 687.5
PAB = 562.5
What is then the relative importance of each factor?
The Variance of the sample can be computed using the given formula:
∑41(𝑦𝑦𝑖𝑖 − 𝑦𝑦�)2 4𝑃𝑃𝐴𝐴2 + 4𝑃𝑃𝐵𝐵2 + 4𝑃𝑃𝐴𝐴𝐴𝐴
2
𝜎𝜎𝑦𝑦2 = =
(𝑛𝑛 − 1) 3
CEN 431 Design of an Engineering Experiment 3/4
Where:
n = 22 = 4
𝑦𝑦�= P0
4(187.5)2 4(687.5)2 4(562.5)2
𝜎𝜎𝑦𝑦2 = + + = 46 875 + 630 208 + 421 875 = 1 098 958
3 3 3
The percentage effect of each factor can be computed using the following formula:
4𝑃𝑃𝐴𝐴2
E(%) = ( 3
)/ 𝜎𝜎𝑦𝑦2 x 100
In this case the results are tabulated below:
% Effect
A = (46 875/1 098 958) ×100 = 4%
B 57% (Important effect)
AB 39% (interaction)
Remark: If the number of parameters is large, the number of experiments required to capture all
effects may be excessively large. For example, a 10-factor 2-level experiment requires n = 210 = 1024
to show out all the interactions.
But,i s it possible to reduce the number of experiments without losing information??
C. Fractional Factorial Design:
• It is not possible to resolve all the interactions.
• A fractional factorial design works just well, only if some the higher order
interactions are of least importance.
VII. References
• Anderson, M. J. and Whitcomb, P. J. 2000. DOE Simplified: Practical Tools for
Effective Experimentation, Productivity Inc.Portland, OR.
• Anderson, M. J. and Whitcomb, P. J. 2005. RSM Simplified: Optimizing Processes
Using Response Surface Methods for Design of Experiments. Productivity
Inc.Portland, OR.
• DU W.Y., Furman, B.J., Mourtos, N.J., On the Ability to Design Engineering
Experiments. Lead paper, Proc., 8th UICEE Conf. on Engineering Education, Feb.
2005.
CEN 431 Design of an Engineering Experiment 4/4
