🌎 All English Articles  |  🇯🇵 Japanese Version

Design Validation Testing: Building a Test Plan That Proves the Design Works

Design Validation Testing: Building a Test Plan That Proves the Design Works English

Twenty years of working as a contract mechanical designer inside major manufacturers in Japan has taught me that a design review sign-off and a validated design are two very different things. I have sat in plenty of meetings where a design was declared “done” because it looked correct in CAD, met the drawing tolerances, and survived one informal bench test that somebody ran the week before the meeting. Design validation testing is supposed to be the discipline that closes that gap — the structured process of proving, with evidence, that a design actually satisfies the requirements it was built against, not just the requirements someone remembered when they wrote the test plan. This article covers how I build validation test plans, the mistakes I have made building them badly, and the habits that have kept me from shipping designs that only look finished.

Starting From Requirements, Not From the Test Equipment You Have

The most common failure mode I see in validation planning is starting from what equipment is available in the test lab rather than from what the design actually needs to prove. A lab with a good vibration table will generate a vibration test plan; a lab with a strong environmental chamber will generate a thermal cycling plan. This sounds obvious when stated directly, but I have watched it happen repeatedly, because it is genuinely easier to design a test plan around known capability than to first extract a clean list of requirements from a specification document, a customer contract, and half-remembered verbal commitments made in a kickoff meeting eight months earlier.

On a control enclosure project, I inherited a validation plan that had thorough vibration and thermal testing but no ingress protection testing at all, despite the enclosure being specified for an IP65 rating in the customer contract. The reason was almost administrative — the test lab had excellent shaker tables and environmental chambers but no water spray rig, so the plan had simply grown around what was easy to schedule. We caught the gap during a requirements traceability review, three weeks before the scheduled validation sign-off, and had to scramble to arrange outside lab time for the water ingress test. It passed, fortunately, but the schedule pressure it created was entirely avoidable. Since then, I start every validation plan with a literal traceability matrix — every requirement in the specification gets a row, and every row needs at least one test method assigned to it before I let the plan move forward, regardless of whether we currently have the equipment to run that test.

Building the Traceability Matrix

Requirement Source Requirement Test Method Pass Criteria Status
Customer spec §4.2 Withstand 5-500Hz random vibration, 2G RMS Shaker table, 2 hours per axis No resonance below 200Hz, no fastener loosening Planned
Customer spec §4.5 IP65 ingress rating Water spray test, IEC 60529 No water ingress after 15min Outsourced
Internal DFMEA Bracket fatigue life >1M cycles Servo-hydraulic fatigue rig No crack initiation at 1.2M cycles Planned
Contract Annex B Operating temp -20°C to 60°C Thermal chamber, 10 cycles Full function at temperature extremes Planned
Field complaint history Latch must survive 10,000 open/close cycles Automated cycling fixture No latch failure, no perceptible wear at 12,000 Added post-review

I keep a table close to this shape for every project of meaningful complexity, and the last row is deliberately included because it represents something important: not every requirement comes from a formal specification. On the project this table is loosely drawn from, the 10,000-cycle latch requirement did not exist in any written document — it came from a field complaint history review on a predecessor product, and it only made it into the validation plan because someone on the quality team insisted on cross-referencing past field issues before sign-off. Formal specifications are necessary but not sufficient; validation planning has to also account for the failure history of whatever came before.

Sequencing Tests to Catch Failures Early and Cheaply

The order in which you run validation tests matters more than most schedules acknowledge. I once ran into a case where a project ran an expensive, time-consuming environmental test — full thermal cycling combined with humidity exposure, taking almost three weeks of chamber time — before running a much cheaper and faster dimensional and functional check at room temperature. The environmental test uncovered a functional failure that turned out to be present even at room temperature, meaning the three weeks of chamber time had been spent validating a design that would have failed a ten-minute bench check run first. We had to fix the root cause, which was a bushing clearance issue unrelated to temperature at all, and then repeat the entire three-week environmental cycle from scratch.

After that experience, I started sequencing validation plans deliberately from cheapest-and-fastest to most-expensive-and-slowest, on the logic that any test early in the sequence that reveals a fundamental design flaw saves you from wasting expensive lab time validating a design that needs to change anyway. Room-temperature functional checks and dimensional verification go first. Basic mechanical tests — static load, simple cycling — go second. Environmental and combined-stress tests, which are almost always the most expensive and slowest to schedule, go last, and only after the design has already demonstrated it works under nominal conditions. This ordering has saved real schedule time on multiple projects since, even though it occasionally means a design change discovered late still forces a partial re-test of the cheaper stages too.

Defining Pass Criteria Before You Run the Test, Not After

I have seen validation tests run without a written pass criterion, with the plan being to “look at the results and decide,” and this is one of the more dangerous habits in engineering because it invites motivated reasoning after the fact. If a bracket bends 3mm under a proof load and nobody wrote down what deflection would constitute a failure beforehand, there is a real risk that 3mm gets rationalized as acceptable simply because the schedule cannot absorb a redesign, not because 3mm is actually fine.

On a linkage mechanism I validated for repeatability, we defined the pass criterion in advance as maximum positional deviation under 0.15mm across 5,000 cycles, based on the downstream tolerance stack that the mechanism fed into. When the test came back at 0.19mm, there was no ambiguity or debate — the number failed the pre-defined criterion, full stop, and the conversation moved immediately to root cause rather than to whether 0.19mm was “probably fine.” That clarity only exists because the number was written down, reviewed, and agreed to before a single test cycle ran. I now treat “what number constitutes failure, exactly” as a mandatory field in every test plan document, and I refuse to let a test proceed without it filled in, even under schedule pressure, because a test without a pre-defined pass criterion is not really validation — it is closer to data collection with an argument attached.

Sample Size and the Danger of Validating a Single Unit

Budget and schedule pressure push toward testing the minimum number of prototype units, often just one, and I understand the pressure, but a single unit passing a validation test tells you less than most people assume. Manufacturing variation, assembly variation, and material lot variation all mean that one good unit could be an outlier rather than representative. On a fan motor bracket project, the single prototype unit we validated for vibration resistance passed comfortably, with a healthy margin against the resonance requirement. When we later built five units from first-article production tooling and ran the same test, one of the five failed, because a casting porosity variation in that particular unit shifted its natural frequency enough to fall inside the excitation band. Had we validated only the original single prototype and moved to production, that failure mode would have shipped.

Since then, I push for a minimum of three to five units on any validation test where manufacturing variation could plausibly affect the result, and I have learned to be specific about which tests actually need multiple units and which do not — a pure geometric fit check on a rigid part rarely needs five samples, but any test sensitive to material properties, casting quality, or assembly tolerance stack-up does.

Documenting Failures as Thoroughly as Successes

The last habit I want to highlight is one that took me longer to build than it should have: documenting a failed validation test with the same rigor as a passed one. Early in my career I saw test reports where a failure was noted in a single line — “failed, redesigned, retested, passed” — with no record of the actual failure mode, the conditions under which it occurred, or the root cause analysis that led to the redesign. This is a wasted opportunity, because failure data is often more valuable than pass data for informing the next design. On the fan bracket project mentioned above, the porosity-related failure became a standing item in our casting supplier quality checklist for every subsequent project, specifically because the failure was documented with photographs, CT scan images of the porosity, and a clear causal chain back to the resonance shift. A validation plan that only records what passed is only half a validation plan; the other half is making sure every failure earns a permanent place in the institutional memory of the team, so the same failure mode does not have to be rediscovered on the next design.

コメント

タイトルとURLをコピーしました