Lean Six Sigma Tools
10 September 2026

Design of Experiments in Lean Six Sigma: Moving Beyond One-Factor-at-a-Time Testing

Share on:

Design of experiments (DOE) lets a Lean Six Sigma team test process inputs together and measure how their effects depend on one another. ASQ places DOE in the Improve phase of DMAIC, where teams evaluate solutions and determine which inputs need control. Before committing production time, the sponsor needs to know whether the experiment will support a decision about operating settings.

One-factor-at-a-time testing leaves interactions unresolved

One-factor-at-a-time (OFAT) testing changes an input while holding others fixed, then repeats the process for another input. It leaves interactions unresolved: the effect of one factor depends on another factor’s setting. The NIST/SEMATECH handbook explains this limitation in its guidance, first released in 2003.

In a Lean Six Sigma root cause analysis, an interaction changes what the team needs to establish. Identifying an input that affects the response is only part of the job. The team also needs to establish how that effect changes across combinations of settings. An OFAT result applies to the values held fixed during the test; it does not establish what happens at untested combinations.

A three-factor experiment requires more than eight production slots

NIST’s worked polishing example examines speed and feed alongside depth, each at two levels. Its full factorial contains eight combinations, allowing estimation of main effects and interactions. Repeating those combinations and adding three center-point runs brings the completed design to 19 runs. NIST presents this as a teaching example and claims no company savings.

Design stage Runs at this stage Cumulative runs
All combinations of three two-level factors 8 8
Repeat every combination 8 16
Add center-point runs 3 19
Run allocation in the NIST/SEMATECH polishing example, from “Full factorial example,” 2003 handbook, online edition.

Approving eight production slots would fund only the initial combinations in this DMAIC experiment. Repeat runs need production capacity too, as does subsequent confirmation. The sponsor’s review should separate the count of unique settings from the total production time requested.

Measurement and run order belong in the DOE plan

ASQ’s Measure-phase guidance includes validating the measurement system before establishing process performance. In a DOE proposal, the response variable needs a clear link to the project’s problem statement, with a defined measurement procedure. For measurements affected by differences between operators or repeated readings, that preparation includes Gage R&R.

Run scheduling is part of the experimental design. NIST’s DOE terminology guidance distinguishes three provisions with different purposes:

  • Randomization: assigning treatment combinations and run order through a random schedule.
  • Replication: performing the same treatment combination again to estimate experimental error.
  • Blocking: grouping runs around known nuisance variables, such as operators or raw-material changes, so their effects can be separated from the effects under investigation.

The project’s Black Belt needs to assess whether the production schedule preserves the intended design. In NIST’s polishing example, running settings in a fixed sequence can confuse a depth effect with day-versus-night conditions. A sequence change requested by operations calls for a design review before the runs begin.

Fractional factorial designs exchange runs for assumptions

A fractional factorial uses a selected subset of the full factorial combinations. Fewer runs come with aliasing: an estimated effect can include contributions from other effects. NIST illustrates this problem with a half-fraction in which main effects are confounded with two-factor interactions.

For a Lean Six Sigma process optimization project, the design review needs to identify which effects remain confounded. If the operating decision depends on separating an input effect from an interaction, the proposal needs to explain how the team will distinguish them. A smaller initial experiment is defensible when the sponsor has budgeted for the follow-up runs needed to resolve those effects.

Coating and lubricant selection in a Lean Six Sigma study

A 2022 study in Materials Science Forum applied DOE within DMAIC to investigate die and punch wear. The study reported a preferred combination of an AlCrN-PVD coating with aloe vera oil. That finding supports a coating-and-lubricant choice under the study’s conditions. Applying the same combination to another production process requires testing under that process’s conditions.

Confirmation runs precede the Control-phase handover

A fitted DOE model needs confirmation at the proposed operating settings. NIST recommends at least three confirmation runs at the proposed best settings, even when those settings appeared in the original experiment. The guidance also calls for reproducing the experimental environment and investigating discrepancies between predicted and observed responses.

At the DMAIC handover, the control plan should identify the confirmed settings and their tested operating range. Those settings then become part of standard work, with control charts used to monitor the process, consistent with ASQ’s Control-phase guidance. Acceptance criteria also need to specify who authorizes a setting change outside the tested range and what evidence that change requires.

Sources

enENarAResESfrFRitITnlNL