Assumptions, Conservatism & Engineering Judgement

What Is Being Demonstrated?

This article addresses how the uncertain inputs and simplifications that are inherent in any structural analysis are managed, documented and justified as part of the substantiation argument. Every analysis contains assumptions: idealised geometry, idealised boundary conditions, assumed material properties, assumed load distributions, simplified failure criteria. These assumptions are not errors — they are necessary simplifications that make the analysis tractable. But they are also the places where the analysis may diverge from reality. The discipline of assumptions, conservatism and engineering judgement is the practice of making these simplifications visible, assessing their effect on the conclusion, and supporting them with evidence or reasoning. An analysis whose assumptions are hidden is an analysis that cannot be challenged, cannot be reviewed and cannot be defended.

ENGINEERING JUDGEMENT SHOULD BE VISIBLE, EXPLAINED AND SUPPORTED — NOT HIDDEN INSIDE THE MODEL.

Why Assumptions Matter

The margin of safety reported by an analysis is only as valid as the assumptions behind it. A positive margin calculated with non-conservative assumptions is not a positive margin — it is a false sense of safety. A negative margin calculated with overly conservative assumptions is not a failure — it is a waste of resources. The engineer must understand which assumptions are conservative, which are non-conservative and which are unknown, and must manage each accordingly. This requires that assumptions be identified, recorded, assessed for their effect on the conclusion and supported by evidence or reasoning. An assumption that is not identified cannot be assessed. An assumption that is not assessed cannot be defended. An assumption that is not defended is a gap in the substantiation.

• A margin is only as valid as the assumptions behind it.
• Conservative assumptions produce lower-bound margins; non-conservative assumptions produce upper-bound margins.
• The engineer must know which assumptions are conservative, which are not and which are unknown.
• An unidentified assumption is a hidden risk in the substantiation argument.

Types of Assumption in Structural Analysis

Assumptions in structural analysis arise at every stage of the process. Understanding where they enter and what they affect is the first step in managing them.

The Assumption Register

An assumption register is a controlled document that records every assumption made in an analysis, along with its basis, expected effect, conservative/non-conservative classification, sensitivity requirement and supporting evidence. The register is the engineer's tool for making assumptions visible and manageable. It is also the reviewer's tool for challenging them. An analysis without an assumption register is an analysis whose assumptions are implicit, unexamined and undefendable. The register should be maintained throughout the analysis and presented as part of the substantiation report.

"Conservative" should not be selected casually. Some assumptions are conservative for one failure mode and non-conservative for another. A thicker section is conservative for static strength but may be non-conservative for fatigue. A higher preload is conservative for joint separation but non-conservative for bolt tension. The classification must consider the specific failure mode being assessed.

Conservatism: What It Means and What It Does Not

A conservative assumption is one that tends to make the analysis result worse — higher demand or lower capability — than reality. Conservatism is desirable when it provides a safety margin against uncertainty. However, conservatism has limits. First, an assumption that is conservative for one failure mode may be non-conservative for another. Second, stacking multiple conservative assumptions can produce a result that is so far from reality that it masks the true governing condition and leads to over-design or misallocation of resources. Third, conservatism is not a substitute for understanding — an assumption labelled "conservative" without analysis of its effect is an assertion, not an engineering position. The engineer must know not only that an assumption is conservative, but how conservative it is and whether the conservatism is appropriate for the application.

• Conservatism makes the result worse than reality — it provides margin against uncertainty.
• An assumption conservative for one failure mode may be non-conservative for another.
• Stacked conservatism can mask the true governing condition and lead to misallocation of resources.
• Conservatism without quantification is an assertion, not an engineering position.

Stacked conservatism is not always safe. Multiple conservative assumptions can produce a result that bears no resemblance to the actual structural behaviour, making it impossible to identify the true governing condition or to correlate with test.

Stacked Conservatism

Stacked conservatism occurs when multiple conservative assumptions are layered in the same analysis. Each assumption individually may be reasonable, but their combined effect can produce a margin that is artificially low — or, in some cases, a failure mode prediction that is entirely artefactual. Consider an analysis that assumes minimum material thickness, minimum material strength, maximum load and a conservative failure criterion. Each assumption is individually conservative, but the probability of all four occurring simultaneously may be vanishingly small. The result is a margin that suggests a problem where none exists, leading to unnecessary redesign, mass increase or schedule delay. The engineer must be aware of stacked conservatism and assess whether the combined conservatism is appropriate or excessive. Where the combined conservatism is excessive, the engineer should de-stack the assumptions — use more realistic values for some inputs while retaining conservatism where it is most needed.

Stacked conservatism effect:

  Individual assumptions (each conservative):
    Min thickness  ───┐
    Min strength   ───┤
    Max load       ───┼──► Combined effect: artificially low margin
    Conservative   ───┤      (may not represent any real condition)
    failure model  ───┘

  The combined probability of all conservative conditions
  occurring simultaneously may be very low.

  Result: a margin that suggests a problem where none exists,
  or a failure mode that is artefactual.

Engineering Judgement

Engineering judgement is the reasoned application of experience, knowledge and logic to a situation where complete data or rigorous analysis is not available. It is not guessing. It is not "I think it will be fine." Engineering judgement is a structured argument: "Based on X, Y and Z, I assess that the structure will behave in manner W, and here is my reasoning." Judgement is necessary because not every question can be answered with a full analysis or a test. But judgement must be supported — by reference to similar cases, by bounding analysis, by physical reasoning or by conservative assumptions. A judgement that is unsupported is an opinion, and an opinion is not substantiation. The engineer must distinguish between judgement that is a shortcut (unacceptable) and judgement that is a reasoned position backed by evidence (acceptable).

Engineering judgement is not the absence of analysis — it is reasoned analysis where complete data is not available, supported by evidence, bounding calculations or physical reasoning.

Sensitivity Assessment

Sensitivity assessment examines how the analysis conclusion changes when an assumption is varied. If the margin remains positive when the assumption is varied across its plausible range, the assumption is not sensitive — the conclusion is robust to the uncertainty. If the margin changes sign or becomes critical when the assumption is varied, the assumption is sensitive — the conclusion depends on the assumed value, and the assumption must be supported by evidence or the design must be changed. Sensitivity assessment is the tool that converts "unknown" assumptions into understood assumptions. It is also the tool that identifies which assumptions matter and which do not — allowing the engineer to focus effort on the assumptions that affect the conclusion.

• Identify the assumptions that could affect the conclusion — typically loads, allowables, boundary conditions and failure criteria.
• Vary each assumption across its plausible range and assess the effect on the margin.
• If the margin is robust across the range, the assumption is not sensitive — document and move on.
• If the margin is sensitive, the assumption must be supported by evidence or the design must be changed.
• Sensitivity assessment identifies where to focus effort — not every assumption needs a full investigation.

Sensitivity:  S = Δ(MS) / Δ(assumption)  — the change in margin per unit change in the assumed value

Challenging Assumptions

Assumptions must be challenged — by the engineer who makes them and by the reviewer who examines them. The challenge is not adversarial; it is the engineering process of testing the argument. For each assumption, the engineer and the reviewer should ask: Is the basis stated? Is the expected effect understood? Is the conservative/non-conservative classification correct for the failure mode being assessed? Has a sensitivity assessment been performed where required? Is supporting evidence referenced? An assumption that survives challenge is an assumption that can be defended. An assumption that does not survive challenge is a gap that must be closed — either with evidence, with analysis or with a design change.

• Is the basis for the assumption stated? — Not just the value — why this value was chosen.
• Is the expected effect on the result understood? — Does the assumption increase demand, decrease capability, or change the failure mode?
• Is the conservative/non-conservative classification correct? — Consider the specific failure mode — conservative for one may be non-conservative for another.
• Has sensitivity been assessed where required? — If the assumption could affect the conclusion, sensitivity is required.
• Is supporting evidence referenced? — An assumption with no evidence and no sensitivity is unsupported.
• Has the assumption been challenged? — Both the engineer and the independent reviewer should challenge each assumption.

The Flagship Principle

This article is designated a flagship article because the management of assumptions, conservatism and engineering judgement is a discipline that underpins the credibility of the entire substantiation process. An analysis with a positive margin but hidden assumptions is not substantiation — it is a calculation with unknown validity. An analysis with visible, challenged and supported assumptions is substantiation — even if the margin is small — because the engineer and the reviewer understand exactly what is known, what is assumed and where the residual uncertainty lies. The discipline of assumptions is what separates a defensible engineering argument from a set of numbers.

The management of assumptions is not a bureaucratic overhead — it is the discipline that makes the analysis defensible. An analysis is only as credible as the assumptions behind it, and those assumptions are only credible if they are visible, challenged and supported.

Key Takeaways

• Every analysis contains assumptions — they are necessary simplifications, not errors, but they must be visible.
• An assumption register records every assumption with its basis, effect, classification, sensitivity and evidence.
• Conservatism is desirable but not universal — an assumption conservative for one failure mode may be non-conservative for another.
• Stacked conservatism can mask the true governing condition — the engineer must assess combined conservatism.
• Engineering judgement is reasoned, supported analysis — not guessing, and never a substitute for evidence.
• Sensitivity assessment converts unknown assumptions into understood ones and identifies where to focus effort.
• Assumptions must be challenged — by the engineer and by the reviewer — to ensure they can be defended.