TL;DR: To validate as built vs design you compare the surveyed reality of a structure against its design model in a single, agreed coordinate frame, then quantify the difference point by point or surface by surface and judge it against a stated tolerance. In practice that means capturing the as-built (total station or laser scan), aligning both datasets to common control on GDA2020/MGA2020 and AHD, running a deviation analysis in software such as CloudCompare, PolyWorks or Leica Cyclone, and issuing a pass/fail report. Get the datum and the tolerance right and the rest is arithmetic.
Key takeaways
- Validation is a comparison, not a measurement — its entire value depends on both datasets sitting in the same coordinate frame. Align the as-built and the design model to common site control (GDA2020 / MGA2020 horizontal, AHD vertical) before you compare anything, or every deviation you report is wrong by the registration error.
- Decide the tolerance before you survey, not after. A SAG mill sole plate verified to AS flatness limits needs ±2–3 mm total-station work; a bulk earthworks design check is fine at ±20–50 mm from a DJI photogrammetry run. Capturing tighter than the decision needs wastes money; looser invalidates the result.
- Match the capture method to the tolerance: Leica TS60 / Trimble S9 total stations give ±1–3 mm for discrete points; Leica RTC360 or FARO Focus scanners give ±2–5 mm registered surfaces; Trimble R12i GNSS RTK gives ±15–30 mm; DJI Matrice photogrammetry with ground control gives ±20–50 mm.
- There are two validation modes and they answer different questions — point-to-point (does this anchor-bolt group / pier / nozzle sit where it was designed?) and surface deviation / heat-map (where does this whole shell, slab or batter depart from the model, and by how much?).
- A deviation report with design coordinate, as-built coordinate, the delta in E/N/RL and a pass/fail flag turns "looks about right" into a number an engineer can sign — and catching a 15 mm out-of-position bolt group on paper saves a fabrication rework and return mobilisation that routinely runs AUD $8,000–$25,000.
What "validate as-built vs design" actually means
Validating as-built against design is the act of proving — with surveyed evidence — that what was physically constructed or installed matches what the engineer specified, within an agreed tolerance. The design is the intent: a 3D model, a set of AS 1100 drawings, or a coordinate schedule. The as-built is the reality: where the steel, concrete, bolts and equipment actually ended up. Validation is the disciplined comparison of the two.
On Australian industrial and mining work this comparison is run for three reasons. The first is conformance and handover — proving to the principal contractor, asset owner or certifier that the works sit inside the project's dimensional control specification before sign-off. The second is fit-up assurance — confirming, before a mill, conveyor module or pipe spool ships from the fabricator, that the receiving structure will actually accept it. The third is brownfield design input — verifying an existing asset so a new tie-in is designed against reality rather than a decade-old drawing.
The critical thing to understand is that validation lives or dies on the coordinate frame. A perfectly captured as-built compared against a design model in a different datum, zone or local grid produces deviations that are pure registration error, not real construction error. Get that wrong and you will condemn good steel or pass bad steel.
Step 1: Define the tolerance and acceptance criteria first
Before anyone mobilises, agree in writing what "pass" means. The tolerance drives every downstream decision — equipment, method, station spacing and processing effort.
Typical Australian industrial criteria look like this:
- Machined / rotating equipment (mill sole plates, kiln tyres, crane rail gauge): single-digit millimetres, often ±1–3 mm, frequently referenced to manufacturer datasheets or AS 1418 for crane rails.
- Structural steel and anchor bolts: commonly ±5–10 mm position, set against the project dimensional control spec.
- Civil and concrete (slabs, footings, piers): ±10–25 mm depending on element.
- Bulk earthworks, batters, stockpile design surfaces: ±20–50 mm is usually fit for purpose.
Tip: Write the tolerance as a signed number with a direction convention (e.g. +E is east, +RL is up). "Within 10 mm" is ambiguous; "±10 mm in plan, ±5 mm in level" is auditable. Ambiguous criteria are where validation disputes start.
Step 2: Capture the as-built with the right method
Choose the instrument to the tolerance set in Step 1, not to whatever is in the ute.
| Validation need | Typical equipment | Stated point accuracy |
|---|---|---|
| Discrete points, machined tolerances | Leica TS60, Trimble S9 total station | ±1–3 mm |
| Whole surfaces, complex geometry | Leica RTC360, FARO Focus, Leica P-series scanner | ±2–5 mm (registered) |
| Civil set-out, larger structures | Trimble R12i, Leica GS18 GNSS RTK | ±15–30 mm |
| Earthworks, stockpiles, broad areas | DJI Matrice / Phantom + ground control | ±20–50 mm |
For point-to-point validation of an anchor-bolt group or pier, a total station occupying known control is faster and tighter than scanning. For surface validation of a mill shell, vessel, slab or batter, laser scanning produces the dense point cloud a heat-map analysis needs. UAV photogrammetry — flown under CASA Part 101 by a licensed remote pilot — is the tool for large open areas where ±20–50 mm is acceptable.
Tip: Capture more control ties than you think you need. Registration quality, not raw instrument accuracy, is usually the limiting factor in a validation, and you cannot improve a poorly tied dataset after demobilisation.
Step 3: Align both datasets to a common datum
This is the step most often skipped and most often regretted. The as-built and the design model must share one coordinate frame before any comparison is meaningful.
In Australia that frame is almost always GDA2020 projected to MGA2020 (the correct zone — Zone 50 across the Pilbara and Goldfields, Zone 55 across much of the eastern seaboard) for horizontal position, and the Australian Height Datum (AHD) for level. Two traps recur:
- Legacy datum mismatch. A GDA94 / MGA94 design overlaid on GDA2020 site control shifts everything roughly 1.8 m to the north-east. Transform deliberately; never assume.
- Local plant grid. Many sites run a rotated, offset local grid. The design model is often in that grid while site control is published in MGA. The as-built must carry the transformation that ties local to national, or the two will never line up.
There are two ways to bring the datasets together. Datum-based registration ties the as-built to surveyed control marks whose MGA/AHD values are known — this is the rigorous, defensible method and the only one acceptable for conformance reporting. Best-fit (ICP) alignment mathematically minimises the distance between cloud and model — useful for a quick first look, but it can hide a real systematic offset by averaging it away, so never use best-fit alone for a pass/fail decision.
Tip: Before trusting any brownfield validation, physically confirm the site control still exists and still holds its published value. Marks get bulldozed, bumped and disturbed; the comparison is only as good as the control behind it.
Step 4: Run the deviation analysis
With both datasets in one frame, run the comparison. Common software includes Leica Cyclone and Cyclone 3DR, Trimble RealWorks, InnovMetric PolyWorks, Autodesk ReCap/Navisworks, and the free CloudCompare for cloud-to-cloud and cloud-to-mesh work.
Two analysis modes answer two different questions:
- Point-to-point. Pick the surveyed feature — anchor-bolt centre, pier corner, nozzle face, rail centreline — and report its delta from the design coordinate in E, N and RL. This is the right mode for discrete, dimensioned elements where the engineer needs exact offsets.
- Surface deviation (heat map). Compute the perpendicular distance from every point in the as-built cloud to the nearest face of the design model (or mesh-to-mesh between two surfaces). The output is a colour-coded map — typically green inside tolerance, warming to red where the surface bulges or dishes beyond it. This is the right mode for shells, slabs, vessels, conveyor structures and earthworks design surfaces.
A worked example: a mill foundation as-built shows the anchor-bolt group offset +8 mm E, −4 mm N against a project tolerance of ±10 mm — that passes, the sole plate will fit. The same group at +18 mm E fails, and you have learned, before the mill ships from the fabricator, that the holding-down arrangement needs reworking.
Step 5: Report pass/fail against tolerance
A validation that ends in a colour picture is incomplete. The deliverable an engineer signs is a deviation table plus a clear statement of method, datum and accuracy.
A good conformance report contains:
- The datum and zone used, and the transformation applied to align the datasets.
- The capture method and stated accuracy, so the reader knows the result is trustworthy to (and only to) that tolerance.
- A deviation table: each monitored point or region with design value, as-built value, the delta in E/N/RL, the applied tolerance, and a pass/fail flag.
- The heat map (for surface work) with its colour scale referenced to the tolerance band.
- A plain-language conclusion — conforms, conforms with noted exceptions, or does not conform.
⚠️ Watch out: The most expensive validation error is reporting a deviation to a tighter tolerance than the data supports. Demanding a 2 mm decision from a ±30 mm GNSS or a ±40 mm photogrammetry dataset is a misuse of the survey — you will either reject conforming work or pass non-conforming work, and both cost real money downstream.
Common mistakes when validating as-built vs design
Mistake 1: Comparing across mismatched datums
Overlaying a GDA94 design on GDA2020 control, or a local plant grid on MGA, produces deviations that are pure frame error.
How to avoid: Identify the datum on both datasets and transform deliberately before comparing — never by assumption.
Mistake 2: Relying on best-fit alignment for conformance
ICP best-fit minimises overall distance and can quietly average out a genuine systematic offset, making non-conforming work look acceptable.
How to avoid: Register to surveyed control for any pass/fail decision; reserve best-fit for indicative first looks only.
Mistake 3: Validating an out-of-date design model
If the model has been superseded by a later revision, you are validating against the wrong intent.
How to avoid: Confirm you hold the current "ISSUED FOR CONSTRUCTION" or controlled design revision before you run the comparison.
Frequently asked questions
What is the difference between as-built validation and a simple as-built survey?
An as-built survey records what was constructed. Validation goes one step further: it compares that record against the design and judges the difference against a tolerance, producing a pass/fail outcome. A survey tells you where things are; validation tells you whether that is acceptable.
Which datum should I use to validate as-built vs design in Australia?
Current best practice is GDA2020 projected to MGA2020 in the correct zone for horizontal position, and AHD for level. The essential point is that the as-built and the design model must sit in the same frame — if the design is in a local plant grid, the as-built must be tied to it through surveyed control.
How accurate does the validation need to be?
As accurate as the decision requires, and no tighter. Machined and rotating equipment typically needs ±1–3 mm total-station work; structural steel ±5–10 mm; civil ±10–25 mm; earthworks ±20–50 mm. The capture method is then chosen to suit. Validating tighter than the tolerance wastes money; validating looser invalidates the result.
What software is used to compare a point cloud against a design model?
Common tools are Leica Cyclone 3DR, Trimble RealWorks, InnovMetric PolyWorks, Autodesk ReCap/Navisworks, and the free CloudCompare. They support cloud-to-model surface deviation (heat maps) and point-to-point comparison against design coordinates.
Can I validate as-built against design using drone data?
Yes, for the right tolerance band. UAV photogrammetry or LiDAR flown under CASA Part 101 with ground control delivers ±20–50 mm, which is suitable for earthworks, batters, stockpile design surfaces and broad civil works — but not for machined or structural-steel tolerances, where a total station or terrestrial laser scanner is required.
Get a validation you can sign off on
Validating as-built against design is only as good as the datum it sits on, the tolerance it is judged against, and the surveyor who understands the asset. Industrial Spatial Solutions runs total-station and laser-scan validations across mining, processing and heavy-industrial sites Australia-wide — properly registered to GDA2020/MGA2020 and AHD, analysed in PolyWorks, Cyclone and CloudCompare, and delivered as clear deviation tables and heat maps with a defensible pass/fail conclusion. Whether you need fit-up assurance before a shutdown, conformance for handover, or a verified record before a brownfield tie-in, call us on 0407 057 015 for a fixed-price quote scoped to exactly the engineering decision your data has to support.