Design Of Experiments

# 1 of 8DOE Sequence 1

Determine the following

• Response(s) output(s) to be optimized (consider location and spread)
• Factors - inputs to be varied
• Factor levels - high and low settings for each factor
• Resource constraints - maximum number of runs for all sequential experiments, factor setting limits, run order, etc.
Design Of Experiments

# 2 of 8DOE Sequence 2

Select Initial Experiment Design

Generally, use approx. 25% of resources for first experiment.

• If 25% of resources allows 4 runs, select 4-run design
• If 25% of resources allows 8 runs, select 8-run design
• If 25% of resources allows 16 runs, select 16-run design Design Of Experiments

# 3 of 8DOE Sequence 3

Determine Sample Size For Each Run

• Variables data: determine sample size for location and/or spread effects (see manuals for calculations)
• Attribute data: determine sample size for proportion (see manuals for calculations)

Only use replication for small full factorials. Repetition is generally more resource efficient. Variance as a response requires repetition or replication.

Design Of Experiments

# 4 of 8DOE Sequence 4

Run The Experiment

• Propriety of conduct is key - minimize experimental noise
• Randomize runs
Design Of Experiments

# 5 of 8DOE Sequence 5

Location

Find the MSFE based on the standard deviation of noise effects.

Use repetitions and compute the standard deviation of the repetitions for each run.

Code data as ln(s2) and treat as a location effect.

OR

Leave uncoded. The effect is the big s2 divided by the small s2. Compare to the critical value of F.

Close

Location

Find the MSFE based on the standard deviation of noise effects.

Use repetitions and compute the standard deviation of the repetitions for each run.

Code data as ln(s2) and treat as a location effect.

OR

Leave uncoded. The effect is the big s2 divided by the small s2. Compare to the critical value of F.

Close

Use graphical methods.

• Normal probability plot
• Pareto Chart
Close

Proportion

Code data as sin-1 (P1/2) and treat as a location effect

OR

leave uncoded and find the MSFE for proportion.

Count

Code data as (c + 3/8)1/2 and treat as a location effect.

Close

Analyze The Results

Determine the significance of effects by using the decision tree below.

Design Of Experiments

# 6 of 8DOE Sequence 6

• If no, proceed to step 7
• If yes, go back to step 1 and begin planning the next experiment
Design Of Experiments

# 7 of 8DOE Sequence 7

Check The Model

Perform an analysis of residuals

• Normal distribution
• Average to zero
• Constant variance
• Randomized.
• Perform a pilot run, if desired.

Design Of Experiments

# 8 of 8DOE Sequence 8

Implement The Improvement

• Confirm the benfit
• Communicate results
• Transfer knowledge to like processes
• Transfer knowledge to other locations