Steel model health checks: finding what cannot be fabricated or erected, before cutting starts

Clash detection answers whether elements collide. Health checking answers a different question: can this be made, can it be delivered, can it be lifted. None of that needs geometric intersection, but it is precisely what a drawing hides — a 13 m beam and an 11 m beam look identical on paper, and the difference is that one exceeds stock length and needs a splice.

What is worth checking

• Length beyond stock. Sections are supplied in stock lengths; anything longer needs either a splice or a special order, and both need to be known early.
• Single-piece weight beyond lifting capacity. Two separate limits: the shop crane decides whether a piece can be turned and loaded in the works, the site crane decides whether it can be placed. The numbers differ and both matter.
• Identical marks with different geometry. The classic detailing error and the expensive one: the shop cuts in batches by mark, so two different shapes sharing one mark means half of that batch will not fit.
• Missing section or material. Elements without these are skipped by take-off, procurement and every later check, while looking perfectly normal in 3D.
• Orphaned elements. Connected to nothing and belonging to no assembly, usually a construction line someone forgot to delete or a part never joined up.

How to compare geometry across a shared mark

This rule sounds trivial and has two implementation traps, each of which produces false positives across a whole real model.

First, do not compare section name plus length. A plate 25 thick, 60 wide and 620 long and a plate 25 thick, 620 wide and 60 long are the same plate; the detailing package simply disagreed about which dimension is the section and which is the length. The correct comparison sorts the numbers from the section together with the length and compares them position by position — a geometric fingerprint rather than a name.

Second, allow a tolerance on the mark itself. Plates marked PL25*144 and PL25*145 are usually the same part, with the millimetre coming from rounding in the detailing package. Without a tolerance, a healthy model produces hundreds of findings.

The criteria have to be shop criteria

Stock length, overhead crane capacity and mobile crane capacity are everyday numbers in a steel fabrication shop. But the same IFC usually also carries slabs, walls, foundations, stairs and reinforcement, and concrete columns and beams are the same IFC entity types as steel ones — only the material tells them apart.

Applying steel thresholds to concrete elements produces findings like "this slab exceeds transport limits", in large numbers, drowning the steel elements that actually needed attention. Health checking therefore has to filter down to steel fabrication parts first: excluding fasteners and assemblies themselves, excluding civil element types, and excluding elements whose material reads as concrete. Precast concrete needs its own thresholds, not steel numbers reused.

Turning off the rules that cannot be done accurately

Some rules sound valuable but cannot be evaluated accurately from an IFC — non-stock sections, beam span-to-depth ratio, column slenderness. What their criteria need, namely the shop's own stock list, the actual span and the radius of gyration, is simply not in the model, leaving bounding boxes as a stand-in. In practice that flags every brace and every tie rod as out of range.

Such rules should stay off even when the code for them is written. A rule that misfires across a model damages more than itself: after three findings that are obviously noise, nobody reads the fourth one that was real.