21-03-2023

Design of Experiments

How to solve problems?

Theory
E=mc2 Experience
• Problem with sheer guesswork in
industrial practice

Experimentation

1
 21-03-2023

Strategy of experimentation
• Extensively used strategy in practice is the one-factor-at-a-time (OFAT)
approach.

• Successively varying each factor over its range with the other factors held
constant

Advantages:
• Simple
• Easy graphical displays

Disadvantages:
• Time consuming
• Extra work
• Shortcut - Less than optimal solution

Strategy of experimentation
Maybe OK if:
•If the response is a complicated
function of the factor, (perhaps
multi-modal) which requires many
levels of X to elucidate
• Effect of factor unchanged by level
of other factors

The above are quite unusual!

2
 21-03-2023

Strategy of experimentation
➢ Keeping time constant

Strategy of experimentation
➢ Keeping temperature constant

3
 21-03-2023

Strategy of experimentation
➢Varying time and temp

Strategy of experimentation
Usually:
• Response is smooth (no sharp kinks or inflection points)
• Effect of one factor depends on level of one or more of other factors

OFAT approach fails to consider any possible interaction between the factors
• Interaction is the failure of one factor to produce the same effect on the
response at different levels of another factor.

4
 21-03-2023

Analysis of historical data

•Wealth of data already available about the

process of interest (computer databases)
•Certainly much more data available from
the logs than could ever be collected in an
experimental program (Excellent precision)
Why experiment?

Correlation ≠ Causation
• A significant correlation between the
response and an independent variable does
not prove causation

Winning a cricket match

Vs
Rain

5
 21-03-2023

Correlated Factors
• Overtime, variation in the factors of interest,
often they vary together
• Impossible to sort out which of the factors, if
any, is having an impact on the response

Shelf life

Humidity, or
Temperature?

Correlated Factors

Why tasty?
Sugar or Syrup?

6
 21-03-2023

No Randomisation = Bias
• No randomization in the variations of the factors
with time, biases to influence the results

One month old bike

New petrol type
Improved mileage
Effect could be due to

Tight Control of factors

• Independent variable when tightly controlled, there will not be enough variation in
that factor to get a measurable effect on the response.

Will the dog bite?

Variation insufficient for effect on response

7
 21-03-2023

Incomplete data

•Some variables not recorded/

Incorrectly recorded
• Or few variables are recorded with
being able to control the process

Hence:

Historical data can never

replace an experimental
program

• But can be used for predictive purposes

8
 21-03-2023

Improper Experiments

Can have all the above problems!

Solution: Proper experimental
design

Good Design Requirements

•Defined objectives
•Unobscured Effects
- Balanced
•Free of Bias
• Blocking
• Randomization
• Replication
•Variability Estimable
•Better Precision

9
 21-03-2023

Blocking Replication
•Repetition resetting all factors
•Keeping noise factors similar
•Improves precision
•Balance the effect of noise variables

Randomization
•Carry out experiments in random order
•Eliminate bias due to lurking factors

Experimental design
Studying the effects of many factors simultaneously in an efficient way, is the purpose
of experimental design techniques

Correct approach to dealing with several factors is to conduct a factorial experiment

•factors are varied together

•Involves Statistical Design of experiments (DOE)
•Computer software programs are widely used to assist experimenters in
selecting and constructing experimental designs.
•Gives meaningful conclusions from the data.
•Obtain superior results with fewer experiments

10
 21-03-2023

Experimental design
Series of tests are conducted by making purposeful changes to the input variables so
that we may observe some concluded results in the output variable.
• For determining which variable has the significant influence on the output.
• For determining optimum combination of process parameters.

✓ Design of experiments is useful for establishing a statistical control of the process

✓ Can be used to determine influential process variables
✓ Improve manufacturing process.

11