Every bridge model looks convincing once it is covered in coloured contour plots. That is exactly the problem. A finite element analysis will produce deflections, moments and stresses whether or not its assumptions bear any relation to the structure on site.
The errors that cause trouble are rarely exotic. They are ordinary decisions taken early — how a bearing is represented, which element is doing the bending, what has been quietly left out — and then never revisited. By the time the results reach a report, the assumptions have hardened into fact.
Here are five mistakes that come up again and again in bridge work, and the corrections that fix them.
1. Over-stiff supports and bearings
A pinned node restrains rotation. Most bridge bearings do not. Model a pot or elastomeric bearing as a simple pin and you have given the deck a rotational restraint it cannot develop in reality. The support attracts hogging moment, midspan deflection comes out low, and the bending moment diagram looks nothing like the one the deck will actually feel.
The same happens when deck and pier are joined with rigid links, or when every support node is restrained in all three translations just to be safe. Stiffness added by accident is still stiffness as far as the solver is concerned.
Practical corrections:
- Represent elastomeric bearings with a vertical stiffness and a horizontal shear stiffness, and let the rotational restraint come from the bearing itself — usually small.
- For pot bearings, model vertical support plus the horizontal restraint the bearing actually provides: fixed, guided or free, according to the arrangement specified.
- Only restrain rotation where the connection can genuinely deliver it, such as a monolithic deck-to-pier joint.
- Halve and double the bearing stiffness and see how much the results move. If the change is larger than you can justify, get real stiffness data.
Support reactions are a useful tell. If a nominally free bearing is picking up horizontal load, something in the model is holding it.
2. Using the wrong element type
Beam elements are excellent for global behaviour and hopeless at shear lag, cross-section distortion or the transverse bending of a slab between girders. Shell elements capture plate bending but say nothing useful about through-thickness behaviour. Solid elements model everything and cost you dearly in mesh size and post-processing patience.
The question is not which element is best; it is which deformation the analysis needs to capture. A deck slab spanning transversely between two main girders needs shells, or at minimum a grillage with a properly calibrated transverse member. A single beam model will give a sensible global deflection and a fiction for the local effects.
A quick sanity test
For every element type in the model, you should be able to say what its degrees of freedom are and which part of the load path uses them. If you cannot, the model is carrying load in ways you never intended — and you will not spot it from the colour plots.
3. Boundary conditions that are missing, not wrong
Over-stiff supports add restraint. Missing boundary conditions remove it. Both distort the answer.
Common omissions include abutment and pier stiffness replaced by a fixed base at the pile cap; soil springs left out entirely, or given one guessed value with no upper and lower bound; temperature, shrinkage and creep absent from a concrete deck; construction stages collapsed into a single load case; and skew or curved supports modelled as though every bearing behaved identically, when in practice the corner bearings carry very different loads.
Sketch the load path by hand before building anything. Where does the load enter, how does it travel, and where does it leave? Every arrow on that sketch should have a counterpart in the model. Then check the boundary conditions against the bearing schedule and the foundation design, not against what was quickest to type.
Where ground stiffness matters, run the analysis with a stiff set of soil springs and a soft set. The truth usually sits between them, and the spread tells you how much the answer depends on a number nobody knows precisely.
4. Poor meshing
Mesh quality affects results long before it affects run time. Long, thin or badly distorted elements produce unreliable bending stresses. Triangles are fine for filling awkward corners and poor at carrying bending. A mesh that is coarse exactly where the load is applied will under-predict local effects.
Two habits help:
- Refine the mesh until the quantity you actually care about — a deflection, a moment, a reinforcement requirement — stops changing. A stable contour plot is not the same as a converged result.
- Do not report peak stresses at a support or a point load as though they were real. Those peaks are artefacts of a singularity in the model. If the local stress matters, take it into a separate sub-model with boundary conditions taken from the global analysis.
5. Results that were never verified
Software does not check itself, and a model that runs without warnings can still be wrong from the first line of input. Verification is not validation: you are asking whether the model has solved the equations you intended, not whether those equations describe the bridge.
Cheap checks that catch most errors:
- Sum the reactions and compare with the applied load, including self-weight.
- Check units throughout. Mixed millimetres and metres have ruined more models than any other single error.
- Estimate a key deflection by hand. A simply supported span under uniform load should land near 5wL⁴/384EI. If the model is three times higher or lower, find out why before you go any further.
- Look at the deformed shape. Loads should travel sensibly from deck to bearings to supports. Odd local distortions usually mean a missing or duplicated element.
- Compare against an independent check — a grillage, a hand calculation, or a previous job of similar form.
A model that fails these checks is not a conservative model. It is simply wrong, and it will be wrong in whichever direction happens to flatter the design.
Making the model earn its keep
A finite element model is an argument, not an answer. Its value comes from the assumptions you can defend, the checks you have run, and the honesty with which you report what the model does and does not represent. Ten minutes spent sketching the load path and sanity-checking reactions will save more time than a fortnight of arguing about a moment diagram nobody trusts.
Keep a short record
Photo: Daniil Komov / Pexels
