The six things a scanning contractor can hand you, and what each one costs in effort

Rung one: the registered point cloud is a deliverable in its own right

A registered point cloud is the honest product of a scanning day: every setup aligned into one coordinate frame, with a report demonstrating how well they agree. It is not a draft of something better. Engineers measure directly in it, check new steel against it, pull a dimension without mobilising to site, and settle arguments about whether a proposed route fits. For a large share of industrial buyers this is the entire useful deliverable and the ladder stops here.

What makes it usable is coverage and registration quality, not point count. A dense cloud with three setups in a congested area is worse than a sparser cloud with twelve, because the twelve fill the shadows. Insist on the registration report as a contract item — Leica's Cyclone REGISTER 360 reports Bundle Error as the average of cloud-to-cloud and target errors alongside overlap and strength per link, and Trimble's Registration Assist reports project and station average error, overlap and consistency.

The cloud is also the only rung that never becomes obsolete through interpretation. Every rung above it embeds a human decision that can be wrong. The cloud embeds the plant as it stood on the day of capture. Archive it in E57, record the coordinate frame, and keep it independent of whatever software the modelling was done in. That archive is what makes commissioning scan-based services years later possible at all.

The formats, and what each one actually carries

E57 is the format to specify for archive. It is standardised as ASTM E2807, Standard Specification for 3D Imaging Data Exchange, Version 1.0, and holds 3D point data, the attributes attached to those points including colour and intensity, and the 2D imagery captured by the same 3D imaging system, in a combined binary and XML structure with a defined extension mechanism. Major hardware and software vendors write and read it, which is exactly the property an archive needs.

LAS is maintained by the American Society for Photogrammetry and Remote Sensing. Version 1.4 expanded the classification field from 32 to 256 classes, introduced Well Known Text encoding for coordinate reference systems, added support for multiple sensor designation and up to fifteen returns per pulse, and lifted the file size ceiling far beyond the previous 4 GB. It is the natural format for airborne and mobile LiDAR that arrives georeferenced.

The Autodesk pair confuses people constantly. RCS is the scan file holding spatially indexed point data. RCP is a lightweight project file that references one or more RCS files together with scan regions, recorded distances and annotations. Deleting the RCS files and keeping the RCP leaves you with an index pointing at nothing. Trimble's X7 exports TDX, TZF, E57, PTX, RCP, LAS and POD, which is a reasonable picture of what moves between systems in practice.

Meshing is not modelling

A mesh wraps a triangulated surface over the point cloud, and photo-texturing it produces the most persuasive image in the whole ladder. It is also the rung most likely to be oversold. Meshing is an interpolation: where the scanner recorded no points, the algorithm generates surface anyway, and it does so most aggressively on the smooth, low-return, awkwardly angled surfaces where the data was thinnest. The result looks most complete precisely where it is least trustworthy.

Meshes earn their keep for visualisation, volume calculation, surface deviation mapping against a nominal shape, and immersive walkthroughs used for familiarisation and safety briefings. They do not earn their keep as a measurement source for tie-in design, because the surface a designer snaps to may be interpolated rather than observed. Where measurement matters, work in the cloud and use the mesh for context.

One legitimate engineering use stands out: deviation analysis. Comparing a mesh of a vessel or tank shell against its nominal cylinder produces a deviation map that flags out-of-roundness, bulging and settlement across the whole surface rather than at discrete points. That output feeds naturally into damage mechanism mapping and RBI visualisation, where geometry and degradation are read together.

Rung four: the drawings most buyers actually need

2D as-built drawings are where a person first enters the process. Someone slices the cloud, decides which returns represent a pipe and which represent scaffold left standing on the day of the scan, identifies fittings, and draws. Plans, sections, elevations and piping isometrics come out the other end as DWG and PDF, readable by every contractor, inspector and permit office without a viewer licence or a training session.

This rung serves more real industrial needs than the two above it combined. Permit applications, demolition scope definition, tie-in packages, scaffold design, fire protection layout, contractor bid sets and simple record-keeping all consume drawings. None of them consume an intelligent model. When a plant discovers that its 3D plant model sat unopened for two years while the drawing set was printed forty times, this rung is why.

The trap is specifying it loosely. "As-built drawings from the scan" says nothing about which elements get drawn, what size threshold applies, whether supports and small-bore are included, or what tolerance the drawn linework holds against the cloud. That last point is the represented accuracy question, and it needs its own number in the scope — a subject covered in full on scan accuracy and tolerance.

Rung five: what makes an intelligent plant model expensive

An intelligent model is not geometry that looks like a plant. It is geometry in which every component knows what it is. A modeller assigns each element a class, a nominal size, a specification, a material and a connectivity relationship to its neighbours, so that the software understands a 6-inch 150# weld neck flange as a flange rather than as a disc of triangles. That assignment is manual, component by component, and it is the entire cost.

BIMForum's definitions describe the gradient precisely. At LOD 300 the quantity, size, shape, location and orientation of the element can be measured directly from the model without referring to non-modelled notes. LOD 350 adds the parts necessary for coordination with nearby elements — supports and connections. LOD 400 is modelled at sufficient detail and accuracy for fabrication of the represented component. Each step multiplies modelling hours on the same field data.

The model pays for itself where automated downstream use exists: clash detection against new design, isometric extraction, quantity take-off, stress analysis input, or a feed into asset records held in an ERP system. It fails to pay where the output is a picture. Before commissioning rung five, name the software that will consume the intelligence. If no such software is named, rung four is the correct purchase.

LOD and LOA measure two different things

Level of Development and Level of Accuracy are routinely used as though interchangeable, and they are orthogonal. LOD, from the AIA schema interpreted by BIMForum, describes how completely and specifically an element is modelled. LOA, from USIBD Document C120, describes how closely the geometry corresponds to physical reality. A beautifully developed LOD 400 assembly placed 40 mm from where the steel actually stands satisfies one and fails the other completely.

USIBD's bands are explicit and stated at the 95 percent confidence level: LOA10 has a lower range of 5 cm, LOA20 spans 5 cm to 15 mm, LOA30 spans 15 mm to 5 mm, LOA40 spans 5 mm to 1 mm, and LOA50 spans 1 mm to zero. The framework draws on DIN 18710, which prescribes five accuracy levels in terms of standard deviation. Version 3.1 was released in 2025; the structure has held since Version 2.0 in 2016.

USIBD also splits accuracy into two halves that a scope must address separately. Measured Accuracy is the standard deviation achieved by the field measurements themselves. Represented Accuracy is the standard deviation achieved once that data is processed into linework or a model. As the specification puts it, error will always be introduced when measured data is represented as some other form of deliverable — so specify a level for each.

Rung six: the digital twin is a process, not a file

Everything through rung five is a snapshot. The moment a spool is replaced, a platform is extended or a valve is relocated, the deliverable begins diverging from the plant, silently and without any indicator that it has done so. Plants routinely discover this three years later when a model they trusted turns out to describe a configuration that no longer exists in the field. The scan did not fail; nobody owned it.

A maintained digital twin closes that gap with two commitments. First, re-capture on change — new scans after every modification, registered into the same frame as the original. Second, data attachment — inspection results, thickness readings, corrosion rates and RBI outcomes bound to the geometry they were taken on, so the model answers questions about condition rather than only about shape. That is what enables thickness trending against real asset geometry instead of against a spreadsheet of CML labels.

This is also the rung where the comparison stops being with other scanning vendors and starts being with the systems a plant already runs. Understanding where a twin sits relative to an inspection data management system prevents buying a second home for data that already has one. The twin adds geometry and spatial reasoning; the IDMS holds the inspection record. They belong together.

Writing the deliverable into a scope you can enforce

Six clauses turn a vague enquiry into comparable bids. Name the rung explicitly, using the ladder above rather than a marketing phrase. State the archive format and require E57 alongside any proprietary deliverable. State the coordinate frame and its origin. State the Measured Accuracy required and the Represented Accuracy required, as separate USIBD LOA bands. State the coverage, including which elevations must be complete. State the verification method to be applied at handover.

Then specify what accompanies the data. The registration report belongs in the deliverable list, not in a vendor's private quality file. So does a coverage map showing setup positions, and a statement of what was inaccessible on the day — because the areas nobody could reach are the areas that will be discovered missing at the worst moment. Vendors who supply all three as standard are the ones who expect to be checked.

Atlantis builds these scopes with the asset and the downstream use in front of us, and delivers whichever rung the use actually requires rather than the highest one available. Capture, registration, drafting, modelling and the maintained data layer are all in scope, sized to the job. Affordable. Accessible. Fully customizable. Send an asset list and the intended use through contact for a quote or a working demonstration.

Request a consultation