Anyone who has sat through a project review after an expensive trial knows the uncomfortable moment when someone asks, “What did we actually learn?”
The equipment was occupied, people waited, material was consumed, and yet the answer may still be that another experiment is needed before anyone can choose a process setting with confidence.
That moment exposes a weakness in how experimental efficiency is often judged. Teams naturally count runs because runs are visible, but the deeper cost comes from the questions left unanswered about interactions, nonlinear behavior, and the operating window. A compact screening design can look economical on the planning slide, then become the costliest option when it must be followed by another study before the process can move forward.
A paper by Peter Goos, José Núñez Ares, Mohammed Saif Ismail Hameed, and Maria Lanzerath offers a compelling alternative. Goos and Núñez Ares co-founded Effex, whose experimental design software is now available as Minitab DOE by Effex, and their published study shows how a 24-run mixed-level orthogonal minimally aliased response surface, or OMARS, design was used to study eight factors affecting the production of a resin for a vascular stent adhesive.
Essentially, it shows how one carefully planned experiment can replace two separate rounds of testing, revealing both what matters and which process settings are most likely to work.
The stakes made experimental efficiency more than a statistical concern. At W. L. Gore & Associates, one experimental resin batch occupied a full-scale production unit for about 40 hours and cost in the five-figure range. With no more than one experimental run possible each week, the usual sequence of screening first and response-surface optimization later could consume months before the team arrived at a usable process window.
A well-selected OMARS design can identify influential factors while also providing the information needed to study interactions, curvature, and promising operating settings.
OMARS designs occupy the useful space between traditional screening designs and larger response-surface experiments. Their structure keeps main effects orthogonal to one another and to second-order effects, helping engineers separate the direct influence of a factor from the interactions and curves that may also affect the response. Mixed-level OMARS designs add flexibility because factors can be studied at either two or three levels, depending on whether nonlinear behavior is considered plausible.
For the stent adhesive experiment, process experts and statisticians selected eight quantitative factors covering polymerization, coagulation and washing, and drying. Five factors were assigned three levels because prior knowledge suggested that curvature might matter, while three were studied at two levels. The team also ranked possible interactions by priority, using years of process knowledge to focus the design without assuming that unexpected effects were impossible.
Minitab DOE by Effex provided 81 candidate 24-run OMARS designs with the required factor structure. The researchers compared them using power, aliasing, and projection properties, eventually selecting a design with no perfect aliasing among two-factor interactions.
The selected design also compared favorably with 22-run and 26-run definitive screening designs. It offered greater power for five of the seven high- or medium-priority interactions, while its power for four quadratic effects was 0.48, compared with 0.27 and 0.29 for the benchmark designs.
The practical lesson is that run count alone does not measure experimental value. What matters is whether the runs appearing in the experimental designable engineers to answer their questions and contribute significantly to determining the next process decision.
The OMARS design gave engineers a practical operating direction and enough confidence to begin robustness testing without running a separate optimization experiment first.
The experiment tracked melt flow index after polymerization and residual impurity after the complete process. The analysis pointed toward lower amounts of two polymerization chemicals to keep melt flow index within specification, along with less acid, more base, lower coagulation and washing temperatures, and more washing cycles to reduce impurity. The selected models also suggested a lower drying temperature.
Read from left to right, the chart shows that three different modeling approaches for the residual impurity may disagree about the shape of individual effects, yet still converge on nearly the same settings, which is the agreement a team needs before committing to the next phase.
To reach that conclusion, the team compared many plausible models rather than trusting a single equation. Engineers also rejected one statistically prominent interaction because it did not make physical sense, since a useful DOE result must survive both mathematical scrutiny and process knowledge. Uncertainty remained visible, but it did not prevent the group from acting.
The process engineers ultimately identified a setting with a high probability of meeting customer specifications and internal process requirements. That setting moved into a robustness study spanning different days, shifts, and raw-material lots, creating evidence for process capability and future lot-release planning.
This progression matters because a successful experiment should do more than explain historical results. It should help a cross-functional team agree on what to do next and provide a defensible path toward reliable production.
For teams working with costly batch processes, OMARS designs provide a key to successful affordable experimentation. The designs’ value comes from turning a limited number of experimental opportunities into a clear operating decision, which shortens the route from development to a process that can consistently deliver what customers require.