3D Scanning Services — LiDAR, Photogrammetry & Drone-Based Reality Capture

In more detail

Atlantis captures process plant, storage tanks, pressure vessels, piping, steel structures and whole sites using terrestrial laser scanning, drone and handheld photogrammetry, and mobile SLAM mapping. Output climbs a ladder: registered point cloud, mesh, as-built CAD, intelligent 3D model, then a maintained digital twin. Accuracy is set by registration control and the tolerance you specify, not by the scanner datasheet.

Scope a capture job from the decision it has to support and the rest of the specification writes itself. A clash check against a proposed tie-in tolerates centimetres; dimensional verification of a fabricated spool does not. The USIBD Level of Accuracy specification bands this explicitly, from LOA 10 at 50 mm through LOA 50 at 1 mm, and pairing an LOA band with a BIMForum Level of Development band separates how well the site was measured from how much of it was modelled. Those are two different costs and two different failure modes. A scanner rated at 2 mm at 10 m, the published figure for the FARO Focus Premium, still yields a wrong cloud when the target network is weak and registration error goes unreported. Ask for achieved registration residuals inside the deliverable, not the instrument datasheet, and name the exchange format: ASTM E2807 E57 travels between every platform.

The asset classes we capture, and what each is really asking for

Process units arrive as a geometry problem with a schedule attached — a tie-in that has to fit, a module that has to land, a scaffold plan priced before the shutdown starts. Storage tanks arrive as a code problem: shell verticality, roundness, bottom profile and settlement, evaluated against API 653 Annex B. Pressure vessels and piping arrive as dimensional control, where nozzle positions and spool lengths stopped matching the isometrics several unrecorded modifications ago. Steel structures, jetties and civil works arrive as a records problem, because the drawing set is lost, superseded or quietly wrong.

Each framing changes what gets captured and how densely. A tank settlement survey needs measured elevations on a defined layout — API 653 Annex B sets the minimum number of survey points at N = D/10, rounded up to an even whole number, with spacing no greater than 31.42 feet. A tie-in scan needs one small volume captured tightly and everything around it captured loosely. Scanning both at the same density is how a capture budget gets consumed by data that no decision ever touches.

We also capture whole sites where the deliverable is navigability rather than measurement: a walkthrough that lets an engineer three time zones away stand inside the unit, click a valve and read its tag. That is a legitimate deliverable with a legitimate and much looser accuracy requirement. Naming which of these four framings applies is the first question in any scoping call, because it sets the instrument, the station count, the registration effort and the deliverable format before anything else is decided.

Four capture methods, and the condition that selects each

Terrestrial laser scanning is the default for process plant, tanks, vessels and confined interiors. It produces ranged measurements rather than inferred ones, and it works in the poor, mixed lighting industrial interiors actually have. FARO publishes 2 mm at 10 m and 3.5 mm at 25 m for the Focus Premium, with a ±1 mm ranging error. Those are instrument figures, not project figures. What reaches your deliverable is that number degraded by however well the individual scan positions were tied together, which is a field discipline rather than a purchase decision.

Photogrammetry, handheld or drone-borne, wins where colour and surface texture carry the information: coating breakdown, corrosion mapping, weld cap profile, forensic and heritage documentation. Drone capture opens flare stacks, tank roofs, jetty undersides, cooling towers and bridge soffits without scaffold or rope access, and the access saving routinely exceeds the cost of the capture itself. The trade is that photogrammetric geometry is reconstructed from imagery and inherits every weakness in the imagery — texture-poor surfaces, specular steel and uniform coatings all degrade it.

Mobile and handheld SLAM mapping trades accuracy for speed across large, cluttered, walkable areas: warehouses, cable basements, mine workings, long pipe racks. It is right where coverage beats millimetres and wrong where a fabricated part depends on the result. Most real industrial jobs blend two methods on a single control network — terrestrial scanning for the units that drive dimensions, drone or SLAM for everything that only needs to exist. Laser scanning versus photogrammetry for industrial work sets out where each one degrades first.

The fourth method is to use what already exists. Where the isometrics, the BIM model or a prior campaign are trustworthy and the decision at hand sits comfortably inside their tolerance, there is nothing to buy. Verifying the existing record against a handful of checked dimensions costs a fraction of a capture campaign and closes a surprising number of enquiries before they become projects. When that is the answer, we say so.

The deliverable ladder, and picking the rung you will actually use

Reality capture is sold as a single purchase and consumed as a ladder. Each rung is built from the one below it, each costs more than the one below it, and each answers a different question. The rung you buy should be the highest one a real decision needs, and no higher. Buying the top rung across an entire site is the most reliable way to spend a capture budget on geometry nobody ever opens, and the overspend is invisible until the model has sat unused for a year.

The move that matters is applying a modelling level per system rather than across the site. A tie-in area justifies object-level modelling with fittings, bolt circles and support steel resolved; the adjacent pipe rack rarely does. Requesting the same level everywhere multiplies modelling hours without changing a single decision, because modelling — not scanning — is where the hours sit once the crew has left site and the invoice starts growing.

The conversion from cloud to a usable engineering deliverable is a separate craft with its own failure modes: over-modelled geometry, objects that look correct but carry no dimensional fidelity, and models delivered with no statement of the tolerance they were drawn to. Point cloud to CAD and as-built deliverables covers how that conversion is specified, checked and accepted, including the acceptance tests worth writing into a purchase order.

Accuracy is a specification you write, not a property you buy

Accuracy in reality capture is not a property of the scanner. It is a specification you write and a provider proves. The USIBD Level of Accuracy specification exists precisely to make it writable: LOA 10 at 50 mm, LOA 30 at 15 mm, LOA 40 at 5 mm and LOA 50 at 1 mm, with the bands drawn from the tolerance ranges in DIN 18710. Version 3.1, published in 2025, added guidance on expressing tolerance as a standard deviation rather than a single figure.

USIBD also separates measured accuracy — how well the site was captured — from represented accuracy, how well the model or drawing derived from it reflects that capture. The two degrade independently. A cloud captured to LOA 40 and modelled carelessly delivers an LOA 20 answer, and nothing in the deliverable warns you. Specify both bands, and require the represented accuracy to be stated on the drawing sheet or carried in the model metadata where a later user will find it.

The number that actually governs the result is registration residual: how tightly individual scan positions were tied to each other and to site control. Published work on bolted-target registration reports mean residuals around 2.3 mm across a target network. Ask any provider what residual they achieved on your job, not what the instrument is rated for. Scan accuracy and tolerance on an industrial plant works through how the two relate and where error enters the chain.

Tanks and vessels: the geometry the codes actually name

Tank and vessel work is where scanning stops being a convenience and becomes the evidence base for a code decision. API 653 Annex B evaluates settlement from measured elevations around the shell-to-bottom weld, with the minimum number of survey points set at N = D/10 rounded up to an even whole number and spacing capped at 31.42 feet. A scan produces those elevations, plus every point between them, from one occupation rather than a manual traverse around a tank in service.

The same capture yields shell verticality, out-of-roundness, bottom profile and local distortion from a single dataset. API 653 places the out-of-plumbness limit for a reconstructed tank at 1/100 of total shell height, capped at 5 inches; an existing tank exceeding it requires an evaluation of suitability for continued service rather than an automatic repair. A scan gives that evaluation a dense measured input instead of a handful of spot checks taken where access happened to be easy.

The decisive advantage is campaign-to-campaign comparability. Two scans tied to the same control network produce a deviation map, which is evidence; two manual surveys with different reference points produce two opinions. That comparability is only real when the control network is re-established, so specify permanent monuments at the first campaign. Where the result then needs an engineering disposition rather than a repair, that work sits with NDT and integrity consulting.

Writing a scope of work a scanner can price and an auditor can check

A scope of work that a scanning contractor can price accurately and an auditor can check later contains seven things, and most published scopes contain two. State the purpose — the specific decision the capture supports. State the accuracy band, referenced to USIBD LOA. State the modelling level, referenced to BIMForum LOD, per system rather than site-wide. State the coverage boundary in plain language, including what is deliberately excluded and why.

Then state the four things that get argued about after handover. Require the control and registration report, including achieved residuals and the coordinate system. Name the deliverable formats explicitly — ASTM E2807 E57 for the exchange cloud, plus whichever native format your engineering team actually consumes. Set the handover medium and the retention period, because clouds are large and providers archive them on their own schedule. Finally, agree revisit terms before mobilisation, not after somebody finds a missed area.

Two clauses are worth adding on any brownfield site. First, require the scan positions to be recorded and delivered, so a later campaign can reoccupy them and produce a comparable result. Second, require a stated tolerance on any drawing produced from the cloud, printed on the sheet itself. Drawings without a tolerance statement become authoritative by accident, and three years later somebody fabricates against a line nobody ever claimed was accurate.

Where capture stops and inspection starts

A scan records surfaces. It measures where material is, never how much of it remains or whether it is sound. Wall thickness, subsurface flaws, cracks, laminations, hardness and metallurgy are all invisible to reality capture, and no increase in point density changes that. Geometry and condition are separate questions, answered by separate work, delivered on separate schedules — and confusing the two is the single most common misunderstanding on a first scanning enquiry.

The two are complementary rather than alternative. The capture tells you where a corrosion monitoring location physically sits and what geometry surrounds it; the examination tells you what the wall thickness is at that point. Both are needed for an integrity decision, and the record that survives an audit is the one where each result is traceable to a place and a procedure. Where the existing inspection records are the weak link rather than the geometry, NDT report validation is the faster fix.

From geometry to a twin that stays true

Geometry is the cheap half of a digital twin. The expensive half is binding condition data to it: reconciling the corrosion monitoring location register, the equipment tag list and the inspection history to real coordinates on a real model. Programmes that capture first and reconcile later capture at the wrong resolution, because the resolution requirement comes from how precisely a CML has to be located, not from how detailed the model looks in a demo.

Do it in the other order. Establish which units carry condition data worth binding and at what location resolution, then scan those units to that requirement and capture everything else at walkthrough fidelity. The result is a twin a reliability engineer opens weekly rather than a showpiece opened once at handover. What a digital twin actually does covers the platform side and the data model behind it.

A twin also ages. Every turnaround changes geometry, and a twin that is never re-captured drifts silently away from the plant until somebody trusts it once too often. Build the re-capture interval into the programme at the start, scoped to the units that actually change. If you are still deciding whether a twin or an inspection data management system is the right container for the data, digital twin versus IDMS sets out the difference.

Sectors served, and what each asks for first

Refining and petrochemical sites ask for tie-in geometry, turnaround access planning and tank programmes first. Terminals and tank farms ask for settlement, verticality and volumetric calibration. Power generation asks for turbine hall access, boiler and duct geometry, and outage planning. Marine and shipyard work asks for hull form, block fit-up and retrofit clearance in spaces where no drawing survived the last conversion, which is most of them.

Aerospace and defence work asks for tooling, fixture and jig verification against nominal, at tolerances an order tighter than plant work — which changes the instrument rather than the workflow. Fabrication shops and modular yards ask for pre-shipment dimensional control, because a module that does not fit on arrival costs more than every scan on the project combined. Mining and bulk handling asks for stockpile volumes, conveyor alignment and structural deformation over time.

What differs between sectors is the tolerance and the reason, not the pipeline: capture, register, report residuals, model to a stated level, hand over in an open format. Tell us the decision waiting on the geometry and the date it is waiting for, and the specification writes itself. Talk to us about a capture scope — demo and quote on request, and affordable, accessible and fully customizable across every asset class above.

Frequently asked

Can you scan a unit that is running?

Yes. Terrestrial laser scanning and photogrammetry are non-contact and line-of-sight, so capture proceeds around live plant under normal permit conditions. What changes is coverage: insulation, lagging, congestion and no-go zones create occlusions, and the scan records the outside of the insulation rather than the pipe beneath it. Say up front which surfaces must be captured bare, because that drives scaffolding and stripping scope rather than scanning scope.

What determines how long capture takes on site?

Station count, not floor area. Congested process units need many short-range positions to see around obstructions; open tank farms and structural steel need far fewer. Access control, permits, escorting and working-at-height arrangements frequently exceed actual scanning time. A realistic mobilisation plan is built from a walkdown or a marked-up plot plan, which is the first thing we ask for.

Do we need to provide site survey control?

Not always, but the deliverable is stronger when you do. Tying capture to an existing plant grid or to permanent monuments means a later campaign can be compared to this one directly, which is what turns two scans into a deformation record. Without control, the cloud is internally consistent but floats in its own coordinate system, and every future comparison begins with a registration argument nobody wins.

Can you work from a point cloud someone else captured?

Yes, where the registration report and coordinate system come with it. Modelling, drawing production, deviation analysis and twin preparation all run on a third-party cloud in ASTM E2807 E57, RCP or LAS. Without the achieved registration residuals the cloud can be used for visualisation and coordination, but it cannot be defended as a dimensional record if a fabrication decision is challenged later.

What happens when the scan shows the drawings are wrong?

That is the normal outcome on a brownfield site and a large part of the reason for scanning. The deliverable becomes the corrected as-built record, and the discrepancy list is handed over separately so engineering can decide what to update and what to leave. Catching it before fabrication converts a site rework into a shop dimension, which is where the return on a capture campaign usually sits.