BEMSC — British Engineering & Manufacturing Standards Council All articles
Technical Briefing

Where the Thread Runs Thin: Tying Simulation Assumptions to Clause Evidence

BEMSC — British Engineering & Manufacturing Standards Council
Where the Thread Runs Thin: Tying Simulation Assumptions to Clause Evidence

Photo: Stanley 2321, CC BY-SA 4.0, via Wikimedia Commons

The appeal of a digital twin is straightforward: fewer physical iterations, faster convergence on a design, and a model that can be interrogated long after the prototype has been scrapped. For many UK manufacturers the commercial case is now settled. The standards case is less so.

A conformity record built on physical testing has a certain blunt integrity. There is a specimen, a rig, a date, a laboratory and a signed report. A record built on simulation has a model, a set of assumptions, a solver version, an input deck and an output. Each of these is defensible. Together they form a chain that is only as strong as its least documented link — and in most organisations we encounter, that link is the assumption.

What Compression Costs

Model-based design moves effort upstream. The decisions that once emerged from a test campaign are now made in a requirements meeting, when someone selects a load case, a material model, a boundary condition or a factor. By the time the simulation converges, those choices are buried three layers deep and are rarely revisited.

This is efficient. It is also where standards evidence quietly thins. A clause that requires a particular performance to be demonstrated does not, in most cases, prescribe simulation; it prescribes an outcome. Demonstrating that outcome by model is legitimate, provided the model can be shown to be adequate for the purpose — and adequacy is an argument, not a file format.

Three Questions a Model Cannot Answer for You

The first is scope. Which clauses is this model intended to address, and which are being addressed elsewhere? A model that implicitly covers everything usually covers nothing in a defensible way.

Second is validity. What is the basis for believing the model represents the physical behaviour that the clause cares about? Validation against test is the conventional answer, but it is only meaningful if the test conditions resemble the conditions of use.

Third is margin. Where the model predicts a value close to a limit, what is the basis for the confidence interval, and who signed it off as adequate? Engineering judgement is legitimate here; undocumented judgement is not.

A Clause-Linked Evidence Register

The practical instrument we recommend is a register that sits alongside the design record and does one job: connect each claimed clause to the evidence that supports it. A row per clause, with the following fields, is sufficient for most small and medium-sized programmes.

The register is not a substitute for the technical file. It is an index into it, and its chief virtue is that a reviewer can see within a minute where the thread is supported by test and where it rests on assumption.

Versioning: The Thread Has a Grain

The most common failure we see is not a wrong assumption but an unversioned one. A model is updated in month seven; the assumption list is not. A solver is upgraded; the input deck is retained but the output is regenerated. A material supplier changes a specification; the model carries the old curve.

Three disciplines address most of this. First, treat the assumption list as a controlled document with its own revision history, not as a preamble in a report. Second, record solver and library versions alongside outputs, and retain the environment in which a result was produced. Third, define a change trigger: any alteration to material, geometry, load case or solver that could plausibly affect a clause-linked result should force a re-review of that row in the register, even if the design intent is unchanged.

When the Model and the Test Disagree

Disagreement is information, not embarrassment. The temptation is to adjust the model until it matches the test — a practice that produces agreement and destroys validity. The better response is to record the discrepancy, investigate its likely cause, and document the engineering position adopted. Where the physical test governs because it is closer to the clause requirement, say so. Where the model governs because the test was unrepresentative, say that too, and explain why.

What must not happen is that the discrepancy disappears from the record because the report was rewritten after the fact.

Defending the Thread Under Review

A reviewer, whether a customer, a notified body, an insurer or a technical committee, will typically ask four things: show me the clause, show me the evidence, show me the basis for the evidence, and show me who accepted it. An organisation that can answer those four questions quickly, from a register, is in a considerably stronger position than one that must reconstruct the answer from a project drive.

A Minimum Viable Practice

For members beginning this journey, we suggest starting small: take one product family, one standard, and build the register for the clauses that matter most commercially. Complete it honestly, including the rows that are thin. Then take it to a review and see how it holds.

The digital thread is a genuine advance. It becomes a documentation gap only when nobody has asked, in advance, what holds it together.

All Articles

Related Articles

The Return That Keeps Returning: Service Records as an Early-Warning System

The Return That Keeps Returning: Service Records as an Early-Warning System

What the Invoice Leaves Out: Appraising Conformity Across a Capital Asset's Life

What the Invoice Leaves Out: Appraising Conformity Across a Capital Asset's Life

Assurance at Arm’s Length: A Sub-Tier Conformity Protocol for UK Manufacturers