Methodological errors in structural verification rarely stop a calculation. The solver finishes normally, reports no numerical warning, and can still produce results that look plausible.

This article walks through six common technical mistakes that distort results while the software reports no failure at all.

1. Peak stress from the contour plot goes straight into the check

ASME BPVC Section VIII, Division 2, Part 5 works with categorized stress components. The stress field is decomposed along a stress classification line drawn through the thickness into membrane, bending, and peak parts, and each part is compared against its own allowable. WRC Bulletin 429 describes how to construct that line on a finite element model. The conditions for comparison are strict. Geometry, thickness, loads, point location, coordinate system, surface, stress components, and classification all have to match on both sides.

The scale of the discrepancy shows up in a worked comparison on a nozzle-to-shell junction. The WRC calculation gives 148 MPa. An aligned readout from the FE model gives 154 MPa, a difference of about 4%. The nodal peak at the weld toe in the same model reads 238 MPa, roughly 55% higher. That number does not enter the check until someone establishes what it represents: a converged structural stress, a fatigue quantity, a notch stress, or a numerical artifact.

Stress categorization is routinely confused with nodal averaging. Averaging smooths stress across adjacent elements to produce a continuous contour plot, and it belongs entirely to the post-processor. At a thickness transition or a material interface, that smoothing hides a real stress jump the structure actually has. Comparing the averaged and unaveraged fields separates a physical gradient from a plotting artifact.

2. Nonlinear results get combined like linear ones

Running several hundred load combinations usually relies on linear superposition. Base load cases are solved once, then scaled and summed at the post-processing stage. The method cuts solver time by a large factor, and it holds as long as the stiffness matrix stays constant.

Once contact, plasticity or large-deformation effects enter the model, the response is no longer generally proportional to the applied loads and can depend on the loading path. The dead load result added to the live load result no longer matches the result of both applied together in a single analysis.

The software still performs the addition and reports formal utilization ratios. The resulting stress field has no physical meaning and corresponds to no real state of the structure. In a nonlinear formulation, every governing combination needs its own solver run, which is a different task in computational scale.

3. ASD and LRFD mixed in one project

The two methods agree only over a narrow range. AISC 360 carries both, but the checks run along different paths. ASD compares service-level forces against nominal strength divided by Ω. LRFD compares factored forces against nominal strength multiplied by φ. The agreement between them is calibrated at a live-to-dead load ratio of three.

Away from that point, the two diverge. Under dead load alone, ASD demands about 25% more section capacity. With equal dead and live load, the gap narrows to 7%. At a ratio of five, LRFD becomes the more demanding of the two.

The divergence itself is not an error. Both methods are code-sanctioned, and either one used consistently produces a defensible design. The error comes from carrying a quantity across from one to the other. A beam sized by ASD under equal dead and live load runs at a utilization ratio of 1.00. Substituting the LRFD factored force into that same check, 1.2G + 1.6Q in place of G + Q, pushes the ratio to 1.40 and reports a failure the structure does not have. The reverse substitution, a service-level force checked against φRn, hides a real capacity shortfall.

4. The thickness correction is missing from the fatigue check

An S-N curve is calibrated at a reference thickness. For DNV-RP-C203 that reference is 25 mm, and a thicker plate performs worse at the same stress range: the crack front is longer and the through-thickness stress gradient is shallower.

The correction enters the check as a factor (t_eff/t_ref)ᵏ, with k = 0.20 for the D-curve in air. For a 40 mm plate, the allowable stress range is 9% lower than for 25 mm. At 100 mm it drops by 24%. The effect on calculated life is larger since stress range enters the cycle count to the third power. The 40 mm plate loses roughly a quarter of its fatigue life, and the 100 mm plate around 56%.

Another error sits inside t_eff itself. This is an effective thickness, and for welded joints it follows from the geometry of the connection, including the width of the deposited weld. Substituting the nominal plate thickness from the material list produces a correction that does not match the detail.

5. The check runs against the wrong National Annex

A Eurocode check rests on two documents. The CEN core text sets the form of the checks. The National Annex of the country where the structure is built assigns the numbers: partial factors γM, combination factors ψ, the mapping of buckling curves to section types, and climatic action maps. Values printed in the core text are marked as recommended, and without a National Annex they carry no binding force.

“Verified to Eurocode” with no country named is therefore an undefined statement.

Running the model against another country’s parameter set produces a properly formatted report that does not match the code enforced at the site. The question turns practical wherever a structure is fabricated in one country and installed in another. Offshore modules, port cranes, and bridge spans fall into that category routinely.

6. Assumptions stay outside the report, results stay outside the model

Building a model means accepting a series of assumptions: a joint treated as pinned, a panel edge treated as fixed. Each decision is justified at the moment of the analysis. Months later, it is often unrecoverable from either the model file or the report.

Exporting results to an outside spreadsheet adds a second problem, because the export breaks the link to the model. When the geometry changes, the exported data does not, and decisions get made on stale numbers.

The gap closes where the check is stored with the model. In structural analysis and code-checking software (https://sdcverifier.com/), check parameters attach to the entity the check applies to and print into the report alongside the result. A reviewer sees that the panel edge was taken as fixed and can argue with it. A spreadsheet cell holding the same value with no explanation leaves nothing to argue with.

The cost of a review that never happened shows up in accident investigations. In a UK tower crane collapse, 24 slew ring bolts failed, and metal fatigue was named as the cause. The same bolts had failed on that crane once before. They were replaced, the cause was never established, and the mechanism repeated.

What actually confirms a result

None of the six errors interrupts the software. The calculation completes without warnings, and a standard report comes out at the end.

Full traceability remains the only reliable criterion. The final utilization ratio has to decompose step by step back to its inputs: the load combination, the code parameters, the section properties, and the assumptions taken at the start. A ratio that will not decompose is not technically confirmed.

DISCLAIMER – “Views Expressed Disclaimer – The information provided in this content is intended for general informational purposes only and should not be considered financial, investment, legal, tax, or health advice, nor relied upon as a substitute for professional guidance tailored to your personal circumstances. The opinions expressed are solely those of the author and do not necessarily represent the views of any other individual, organization, agency, employer, or company, including NEO CYMED PUBLISHING LIMITED (operating under the name Cyprus-Mail).