What is design of experiments?
Changing one factor at a time is slow and misses interactions. A designed experiment changes several factors together in a balanced pattern, so every run contributes to every effect estimate. Two-level factorial designs are the workhorse of the Improve phase of DMAIC: they screen many potential causes found with a fishbone diagram or a cause-and-effect matrix, and quantify the vital few.
Running a factorial experiment
- Define the response and how it is measured (check the measurement system first).
- Choose factors and levels far enough apart to see an effect, but within a safe operating window.
- Choose the design: full factorial up to 4 factors; half fraction to screen 4 or 5 factors.
- Run in the randomized order and record any unusual events.
- Analyse the effects, then confirm the best settings with confirmation runs.
DOE example
A pouch sealer was studied with three factors: seal temperature (150 / 170 °C), seal pressure (2 / 4 bar) and dwell time (0.5 / 1.0 s), as a full factorial with two replicates (16 runs). Temperature has the largest effect (+4.2 N/15 mm from low to high), dwell time adds +2.5, and the temperature × dwell time interaction is also significant (+2.4): a longer dwell helps much more at high temperature. Pressure has no significant effect. The best settings to maximize seal strength are 170 °C and 1.0 s, with pressure set where it is cheapest and gentlest on the film; the model predicts about 17 N/15 mm.
Tips for better experiments
- Hold everything else constant and record what you could not control.
- Add center points in a follow-up experiment to check for curvature.
- Start with screening (fractional design), then refine the vital few factors.
- Lock in the result: update the process settings under change control and monitor with a control chart.
How to use this Design of Experiments tool
- Choose factors and levels. List 2 to 5 factors and their low and high settings, with units.
- Choose the design. Pick a full factorial or a half fraction, and the number of replicates.
- Run the experiment. Follow the randomized run order and type or paste the result of each run.
- Find the significant effects. Read the Pareto of standardized effects, then the main effects and interaction plots.
- Pick the best settings. Use the recommended settings for your goal and confirm them with extra runs.
- Export. Export the run sheet, effects and plots to PDF or Excel.
Frequently asked questions
What is design of experiments (DOE)?
Design of experiments is a structured way to change several process factors at the same time, in a planned pattern, so you can estimate the effect of each factor and of their interactions with far fewer runs than changing one factor at a time.
What is a 2-level factorial design?
Each factor is tested at a low and a high level. A full factorial runs every combination: 2² = 4 runs for 2 factors, 2³ = 8 for 3, up to 2⁵ = 32 for 5. It estimates all main effects and interactions, and is the most common design for screening and improvement.
What is a fractional factorial and what is resolution?
A half fraction runs half of the combinations, at the cost of aliasing: some effects cannot be separated. Resolution III designs alias main effects with 2-factor interactions; resolution IV aliases 2-factor interactions with each other; resolution V keeps main effects and 2-factor interactions clear. The tool shows aliased terms together, for example AB + CD.
Why randomize the run order?
Randomizing protects the results from time-related effects such as tool wear, warm-up or a change of material lot. Without it, a drift could look like a factor effect. The tool gives a random run order you can reproduce, and you can re-randomize.
Why replicate?
Replicates estimate the pure experimental error, so the significance of each effect can be tested with regression. Without replicates, the tool uses Lenth's method, which assumes most effects are small. Replicating also makes the conclusions more robust.
What is an interaction?
An interaction means the effect of one factor depends on the level of another. In the interaction plot, the lines are not parallel. When an interaction is significant, choose the levels of the two factors together rather than one at a time.