TL;DR
Tolerance analysis compares a registered cloud against a stated reference and reports deviations as numbers, not colours. The uncertainty budget — ±5mm registered, 2–4 mm RMS interior, 8–12 mm to control — sets the minimum meaningful tolerance. It excels at floor flatness, column plumb, wall bow, and fit-up checks; it cannot resolve movement below its own noise. Reports need a named reference, a bounded region, a fixed colour ramp, and a deviation table.
Engineers do not need a pretty model. They need to know whether a floor is flat enough, a column is plumb enough, a slab is thick enough, and a new assembly will fit the opening that exists. Tolerance analysis is the workflow that answers those questions from scan data, and the quality of the answer depends entirely on how the analysis is bounded.
What tolerance analysis actually is
Three inputs, one output:
- A measured surface — the registered point cloud, ±5mm on our building-scale work.
- A reference — the design model, a nominal plane, a specified slope, or a code limit.
- A tolerance — the allowable deviation, from a specification, a standard, or the engineer's judgement.
The output is a deviation field, plus a table of values at named locations, plus a pass/fail against the stated tolerance. The table is what gets cited in a report; the colour map is for orientation only.
If the deviation being measured is not comfortably larger than the measurement uncertainty, the analysis has no verdict to give.
The uncertainty budget, and why it decides your minimum tolerance
Deviation values inherit the whole chain:
- Registered cloud accuracy: ±5mm
- Interior registration residuals: 2–4 mm RMS; exterior 4–6 mm
- Tie to project or state plane control: 8–12 mm absolute
- Reference model accuracy, if comparing to a model rather than a nominal plane
Comparing a cloud to a *nominal plane* fitted from the cloud itself avoids the control term entirely — this is why floor flatness and column plumb analyses are the most reliable products in this category. Comparing to a *design model* pulls the full 8–12 mm in, because both datasets have to share a frame.
Practical consequence: analyses against tolerances in the tens of millimetres are solid. Analyses against tolerances at or below the registration noise are not analyses, they are noise maps.
What engineers use it for, ranked by how well it works
Floor flatness and levelness. Excellent. The cloud gives full-surface coverage that a straightedge survey can never match, and the reference plane comes from the data. Critical for equipment pads, rail-guided systems, and tenant fit-outs.
Column plumb and member straightness. Excellent. Fit an axis, measure departure, tabulate by grid line.
Wall bow and out-of-plane deviation. Very good, especially on masonry. Historic envelopes are never planar and the deviation drives every glazing and millwork decision downstream.
Fit-up checks for new assemblies. Very good, and the highest-value use in retrofit work. Measure the opening, compare against the fabrication model, and find the interference in the shop instead of on the hoist.
Clearance and code compliance checks. Good. On a 480,000 sq ft campus survey, the federated deliverable flagged 19 non-compliant curb ramps against slope limits — a check that would otherwise require walking every ramp with a smart level.
Small-scale deformation trending. Poor. Movement below the registration uncertainty is invisible. See 3D laser scanning for deformation studies for where instrumentation takes over.
Setting the analysis up so it survives review
- State the reference explicitly. "Best-fit plane through the analysis region", "design model rev C dated…", "nominal slope 1:48". A deviation map without a named reference is uninterpretable.
- Bound the analysis region. Include the slab, exclude the equipment bases and the trench drain. Unbounded regions produce alarming statistics driven by things nobody was asking about.
- Fix the colour ramp to the tolerance. Auto-scaled ramps make compliant structures look like failures. Set the limits to the engineering tolerance and the map becomes readable at a glance.
- Report values, not just colours. A table of deviations at named grid intersections is what an engineer signs. The map orients; the table decides.
- State the measurement uncertainty in the report. Every deviation should be readable alongside the accuracy that produced it.
- Exclude occluded areas. Regions with thin or missing coverage get masked, not interpolated.
Deliverables to specify
- Registered cloud in E57 and RCP, with a registration report
- Deviation maps per analysis region, ramp limits stated
- A tabulated deviation schedule at named locations
- A written statement of reference, tolerance, and uncertainty
- Optional: the reference model, updated to as-built where the analysis found departure
For modelling scope alongside the analysis, LOD should be set per system rather than globally — the reasoning is in LOD 200 vs 300 vs 400, and the cost effect is real: on a 180,000 sq ft conversion, LOD by system cut modelling hours 28%.
What it costs
Capture runs 80,000–120,000 sq ft per day on open interiors, with capture and registration in the $0.05–$0.20/sq ft band. Analysis scope is priced by region and reference complexity rather than by area, since a floor flatness study across a warehouse and a fit-up check on twelve openings are very different amounts of work on the same dataset.
Set against 15–40 RFIs at $1,200–$2,500 each on projects working from unverified geometry, a bounded tolerance analysis on the two or three assemblies that actually carry risk is the cheapest engineering hour on the project.
Discussing a specific tolerance question? Call 614-210-3679, or see Scan to BIM and As-Built Documentation.