BIM progress monitoring: keeping the as-built model current

· Postlight Research

Most projects start with a detailed BIM model and finish with a record model at handover. In between, the model and the building drift apart week by week. This note covers why that happens, how spatial capture from cameras and LiDAR can map what is on site back onto IFC elements without asking anyone to file a report, and which privacy controls that capture needs.

Why the model falls behind the site

Design-office models are maintained by the design office. A Revit or IFC model is updated when a drawing revision is issued, not when a wall goes up. The site runs on a different record: marked-up PDF drawings, photos in messaging groups, daily reports typed at the end of a shift, and the memory of the site engineer who walked the floor that morning.

The gap compounds. A progress claim is checked against photographs rather than the model. A deviation, a column set 30 mm off grid or an opening shifted to clear a service, is found by the next trade that runs into it rather than by the engineer who could have designed around it. At handover, someone rebuilds the as-built model from the paper trail, and the paper trail has holes.

Apps that ask site staff to record progress help when people use them. Use drops when the reporting competes with the work, which is most of the time. A monitoring system that depends on extra reporting inherits that problem.

Capture that rides along with the work

The alternative is to capture the site as a side effect of people moving through it. Three capture modes are common, and each makes a different trade.

Illustrative: capture modes compared. Real accuracy depends on the equipment and the procedure.

Capture mode What it records Strength Limitation
Helmet camera, stereo or 360 Video along the path a person walks Daily coverage with no dedicated survey Depth from images is less precise than laser ranging
Handheld or mobile LiDAR A dense point cloud along a walked path Good geometry for walls, slabs and openings Needs a deliberate walk; drift grows without control points
Terrestrial laser scanner A survey-grade point cloud from fixed stations The most accurate deviation checks Slow, so it is used at milestones, not daily

Daily video from people already on site gives coverage; periodic scans give accuracy. A useful system combines the two: frequent observation of what exists, and precise measurement where a deviation matters.

From pixels to elements

Capture is not monitoring until each observation lands on a model element. The pipeline has four steps.

Reconstruct. Structure from motion and visual SLAM recover the camera path and a 3D reconstruction from video; LiDAR adds measured depth. The output is a point cloud or a set of surfaces in a local coordinate frame.

Register. The reconstruction is aligned to the model's coordinates. Coarse alignment comes from features that are built early, such as grid lines, cores and stair shafts; fine alignment fits the reconstruction to surfaces already built with iterative closest point. Registration error sets the floor under every measurement that follows, so it has to be reported, not hidden.

Recognise. For each element in the model, the system decides what the reconstruction shows at that place: nothing yet, partial work, a finished element, or something the model does not expect. Geometry alone is often ambiguous, since a formwork panel and a finished wall occupy the same space, so appearance in the images carries real weight.

Assign. Each observation is written back against the element it describes, with a state and its evidence.

Illustrative: element states as a monitoring system might record them.

State Meaning Evidence kept
planned In the model, not yet seen on site The last capture that covered the location
in progress Partly seen: formwork, part-built blockwork, first fix The frames and points that show it
built Seen complete and within tolerance Capture date and measured position
deviation Seen, but displaced from its design position Offset in millimetres, its direction, and the source capture

Writing it back into IFC4

Observations are only useful in a format the rest of the project can read. IFC4 already has the structure, so there is no need for a parallel database.

Elements sit in the spatial hierarchy, IfcSite, IfcBuilding, IfcBuildingStorey and IfcSpace, so an observation can be scoped to the room or zone where it was made. Work is described by an IfcWorkSchedule and its IfcTask entries, nested to follow the programme. IfcRelAssignsToProduct links each task to the elements it builds, and each task's IfcTaskTime carries the actual start, the actual finish and a completion ratio. Observed state, measured offset and the identifier of the capture clip behind it can be stored as property sets on the element, so any IFC viewer shows them and any script can query them.

Provenance is the part teams most often skip. Every change of state should name the capture it came from, its date, and that capture's registration error. When a planner disputes a progress figure or a structural engineer questions a deviation, the answer is the clip and the measurement, not a number on a dashboard. Writing each update as a new revision, rather than overwriting the file, gives a model that can be replayed week by week.

Privacy on a filmed site

Filming a workplace affects the people in it, and a monitoring system should treat privacy as a design requirement, not a policy document. The controls that matter in practice:

These controls also make capture easier to adopt. People who know the footage cannot be turned against them are more willing to wear the camera.

What to ask of a progress monitoring system

When you evaluate a system on your own project, ask for the evidence behind a sample of states: pick ten elements marked built and ten marked as deviations, and ask to see the capture and the measurement for each. Ask what the registration error was for the captures behind them. Ask whether the output is an IFC file your team can open, or only a web view. Ask what happens when two captures disagree about the same element, and which one wins. And ask what site staff had to do differently to produce it; the right answer is close to nothing.

Postlight is building this kind of site twin for construction companies: see how it works for builders. For existing buildings, where the starting point is a scan rather than a design model, read point cloud to BIM with AI.