Slipcurve›BIM and 4-D

A clash that becomes a tracked issue against the critical path

Real IFC import, rotation-aware clash detection, exclusion rules that count what they suppressed — and a triaged clash that leaves as BCF 2.1 and comes back as a coordination issue tested against its activity's total float.

The point of a clash run is the issue, not the count

A clash report with 4,000 rows is not a deliverable. The reason model coordination tools exist is that a geometric interference becomes a tracked issue with an owner and a due date, and every hour of the process after detection is about triage.

So the clash manager here carries status, assignee, note and a trade-group filter, and a triaged clash promotes into the coordination issue log — where it joins the risk and issues hub, the related-items panel, the alert feed and the reports. And because the promoted row carries the linked activity, delay propagation runs on it: the issue is tested against that activity's total float and published as a schedule alert if it will move the finish.

Nothing else in this market connects a geometric interference to the critical path.

Promotion reports how many rows it skipped and never overwrites a coordinator's later edits, because a second run of the same clash test must not undo the triage the first one produced.

BCF 2.1 export from a clash test, written properly

The BIM Collaboration Format is how an issue leaves one tool and opens in another, and it is what an open BIM coordination workflow runs on once the models are federated. Export here produces BCF 2.1 — one topic per clash, with a markup file, a viewpoint carrying a perspective camera framed on the clash volume, and the film frame as its snapshot — so a clash test run here opens as a tracked issue in Navisworks, Solibri or BIMcollab, beside the topics those tools raised themselves.

Four things in that writer fail silently if they are wrong, and each is pinned by a known-answer test:

  • The status words must be words the issue hub classifies, the issues module keeps, and the drawings module's own dropdown offers. The test asks the hub itself rather than restating a copy of the list.
  • The duplicate guard on the clash key, so a second export does not produce a second topic for the same interference.
  • The viewpoint camera, in closed form: a two-metre cube at the origin puts the camera at 3.5, 3.5, 3.5. A camera computed slightly wrong produces a viewpoint that opens somewhere near the clash, which is worse than one that obviously fails.
  • The topic identifier, derived rather than random, so re-exporting the same clash produces the same topic.

And one rule that is about honesty rather than correctness: a topic's IfcGuid is written only for an element that really came from an IFC file. A fabricated GlobalId is one the receiving tool fails to resolve, and it says nothing at all about which element is meant.

IFC import that refuses to invent a volume

The importer reads STEP text directly into elements, storeys, property sets, world placements and measured bounding volumes, applying the file's own declared length unit. Elements link to schedule activities by type, by storey, by property set, or by hand.

Anything whose extent cannot be derived is named and excluded — never given a zero-size box.

A zero-size box is the tempting shortcut, and it is exactly the wrong one: it clashes with nothing, so it disappears from every report while still appearing in the element count. The importer's coverage across the representation types a real exporter emits is measured and recorded in the product's own documentation rather than asserted.

One implementation decision worth stating, because it is the kind that is usually hidden: the well-known WebAssembly IFC library is not used, and the reason was measured rather than assumed. A .wasm file cannot be fetched at all from a page opened from a file path; the only working route re-encodes it to base64, at which point the shipped bytes stop being the vendored bytes and the audit trail breaks; and its licence is MPL-2.0 rather than MIT.

The clash test itself

Oriented bounding boxes with a separating-axis test — rotation-aware and z-aware, so a rotated duct and a rotated beam are tested as the shapes they are rather than as their axis-aligned envelopes. Cranes and booms carry per-part volumes rather than one envelope around the whole machine, which is the difference between a useful crane clash and a permanent false positive.

Every clash carries a measured penetration depth and a z-range, which is what makes triage possible: a two-millimetre touch and a 300-millimetre interference are different problems, and a tool that reports both as "clash" has moved the work rather than done it.

A beam and a duct intersecting, with the penetration dimensioned in plan and in elevation Two views of the same pair of elements. In plan a beam 400 wide runs east-west and a duct 700 wide runs north-south across its eastern end; they overlap by 300 millimetres along the beam and by the beam's full 400 millimetre width across it, so the plan penetration is 300. In elevation the duct passes 180 millimetres below the soffit of the beam. The reported penetration is the lesser of the two, 180 millimetres, because a clash is only as deep as its shallowest axis. PLAN IfcBeam · 400 × 600 IfcDuctSegment 700 × 450 300 mm ELEVATION 180 mm soffit of beam REPORTED PENETRATION 180 mm min(plan 300, vertical 180) — the shallowest axis wins hard clash · above the 150 mm reporting floor
Both views are the same pair of elements. The plan penetration is 300 mm and the vertical overlap is 180 mm, and the reported figure is the lesser of the two: a clash is only as deep as its shallowest axis, and reporting the 300 would send a coordinator looking for a problem two-thirds bigger than the one that is there. The 180 clears the reporting floor, which is what stops a 60 mm screed lap arriving in the log as a defect.

Clash tests, rules and exceptions — and what they suppressed

A clash test here is a saved object rather than a one-off sweep: two selections over the imported elements, a geometry mode, a tolerance, its exclusion rules and its exceptions, kept together so next week's run is comparable with this one. The selections are the sets a coordinator actually works in — structure against services, this storey against the storey above, one subcontractor's package against everybody else's.

Exclusion rules are how a false positive stops arriving every week. They run after the narrowphase rather than before it, and that ordering is the whole point: the ledger can then tell you a rule hid so many real interferences, instead of reporting that those pairs were never tested. Every declared rule gets a row even when it suppressed nothing; a rule whose field neither element carries reports unevaluated rather than "no match"; and the ledger foots — pairs tested equals rows reported plus rows suppressed, checked rather than asserted.

A suppressed clash you cannot count is a clash you have not seen.

Four intersection exceptions, the duplicate / inside / overlapping classification and orientation-and-size severity bands are all here. What is refused is an empty answer to a broken question: a selection that matched nothing, or a geometry mode that cannot be honoured, comes back blocked by name rather than as a clean zero. And the volume tolerance never suppresses — a volume figure is trustworthy only on solid geometry, these extents are boxes, so the row is flagged and kept.

4-D sequencing, and the site as part of the network

  • The 4-D film is driven by the schedule: elements appear as their activities are built, so a sequencing error is visible as a wall arriving before the slab under it.
  • Site logistics linked to the plan. A laydown area, a crane position or a hoarding line is an object that links to a schedule task — and delaying that object propagates through the CPM network. No logistics-planning tool in this market connects a laydown area to a float path.
  • Haul times applied honestly. The supply map uses live road data and tags each result with the feed and the time it was measured, on the drawing as well as in the table; a feed that does not answer is named by its reason where the number would have been, in words. Applying a haul time to an activity is max(current, haul-driven) with an undo, so a distant supplier can only ever lengthen a delivery-dependent activity.
  • Reverse-geocoded addresses come from a live lookup and carry that lookup on the row; a row the lookup cannot answer says so instead of carrying an address.

What this does not do

  • A shipped rule library. A clash test carries its own selections, exclusion rules, exceptions and tolerance — but you author them. There is nothing here like Solibri's library of pre-written discipline rulesets to start from, and line and point geometry modes are refused by name rather than approximated.
  • Revit, IFC4x3 and proprietary format import. IFC STEP text is what is read.
  • Sending a marked-up sheet as a PDF. A sheet can be marked up here — notes, callouts, clouds, arrows, stamps and freehand, each carrying its author and its accept or reject — and two issues of one drawing can be overlaid with the changes found and clouded. What does not exist is a flattened or annotated PDF export or a print, so a markup reaches someone who does not open this workspace only through the workbook, never as a PDF.
  • Model federation and version comparison across disciplines. Models are imported; they are not managed.

Related