As-built documentation by laser scanning for brownfield revamp and tie-in projects
The drawings are wrong before the project starts. On a brownfield revamp, a registered point cloud becomes the dimensional record of record: every line, support, foundation and obstruction measured once, then checked against the new design model. Interferences surface in the model, where a designer moves a line in an hour, instead of in the field.
Age is the root cause and it is measurable. The EIA identifies Marathon's Garyville, Louisiana refinery — which came online in 1977 and now runs 617,000 b/cd — as the newest US refinery with significant downstream unit capacity. Nearly every operating US process unit therefore predates the CAD systems its drawings are supposedly held in, and every one has absorbed forty-plus years of modification: a support relocated during a 1994 outage, a conduit run added in 2011, a line rerouted around a new exchanger and never redlined back. NIST put the annual cost of inadequate interoperability across the US capital facilities industry at $15.8 billion on a 2002 basis, excluding residential and transportation infrastructure (GCR 04-867), and found the operations and maintenance phase carrying the largest share. A scan does not fix the drawing archive. It replaces it as the thing you measure from.
Source: EIA, When was the last refinery built in the United States? and Refinery Capacity Report (capacity as of 1 January 2026); NIST GCR 04-867, Cost Analysis of Inadequate Interoperability in the U.S. Capital Facilities Industry (Gallaher, O'Connor, Dettbarn and Gilday, 2004); Construction Industry Institute RS153-1, Field Rework Index, and CII Publication 10-2; USIBD Level of Accuracy Specification v3.1 (January 2025), levels derived from the tolerance ranges in DIN 18710; PFI ES-3, Fabricating Tolerances; ASME B31.3; Leica Geosystems RTC360 datasheet; 29 CFR 1910.146.
| Discovery point | Who resolves it | Correction path | Schedule impact |
|---|---|---|---|
| Design review against the point cloud | Designer at a workstation | Move the line in the model, reissue the isometric | None — absorbed inside design float |
| Clash run against the federated model, pre-IFC | Design lead and piping engineer | Reroute, resize a support, revise before fabrication release | Days, contained inside the engineering schedule |
| Fabrication shop, before shipping | Fabricator raising an RFI | Re-cut, re-fit, reissue — at shop rates, in shop conditions | Weeks, but off the project critical path |
| Field fit-up during the outage | Field crew, on the critical path | Field measurement, field-fit weld, cut and re-prep, or a hot work permit for a modification | Shift-level delay per occurrence, compounding into startup |
| Startup and commissioning | Operations and engineering | Punch list item, temporary support, or a return outage | Carried as deferred work into the next turnaround |
Why a forty-year-old unit has no correct drawing
Start with the age. The EIA identifies Marathon's Garyville, Louisiana refinery, online since 1977 and now running 617,000 b/cd, as the newest US refinery with significant downstream unit capacity. Almost every operating process unit in the country therefore predates modern CAD, and its original documentation was drawn on paper by hand, microfilmed, scanned, and at some point traced into a system that treats the tracing as authoritative. Each of those conversions preserved the drawing and lost the survey.
Then add forty years of modification. A support relocated during a 1994 outage to clear a new pump. A conduit run added in 2011 by an electrical contractor working a separate scope. A line rerouted around an exchanger that was itself a replacement for a different exchanger. Each change was correct on the day and each redline had a defensible reason not to make it back into the master. The cumulative result is an archive that is individually plausible and collectively unreliable.
The errors are not randomly distributed, which is what makes this dangerous rather than merely annoying. They concentrate in congested areas, because congested areas attract modification, and revamp scopes concentrate in exactly those areas for exactly the same reason. The drawing is most likely to be wrong precisely where the project needs it to be right.
The tie-in is where the money is
Every revamp reduces to a set of tie-ins. Cutting into an existing header commits the project to a flange face position, a branch angle, a bolt hole orientation, a support elevation and a clearance envelope — all decided months ahead of the outage in which they get tested. Get them right and the spool drops in. Get one wrong and a crew stands in front of a piece of steel that does not fit, holding a crane and a critical path.
A registered point cloud changes what those commitments are made from. Dimensions extracted from measured data carry a stated accuracy, a capture date and evidence that survives an argument. Dimensions taken from a legacy isometric carry a revision number and an assumption. PFI ES-3 sets the fabricating tolerance band — ±1/8 in (3.0 mm) for pipe 10 in and under, ±3/16 in for 12 through 24 in, ±1/4 in over 24 in through 36 in — and a spool built precisely to tolerance against a wrong datum is still a spool that does not fit.
The second-order effect is the field-fit allowance. Without measured data, fabricators leave generous field-fit lengths on everything and the crew grinds to suit, which converts shop hours into field hours at the worst exchange rate on the project. With measured data, field fit is reserved for the runs the cloud genuinely could not see. That reallocation is where a revamp gets its schedule back.
Clash detection against the new design model
Clash detection between two design models is well understood. Clash detection between a design model and a point cloud is a different problem, because a cloud is millions of discrete points and a naive intersection test flags every one of them separately. The workable approach treats the cloud as a set of recognised objects or as a bounded envelope, so an interference reports as one clash against a pipe rather than four thousand clashes against points.
Two clash classes matter on a revamp. Hard clashes are geometric intersections — new steel occupying the same space as existing steel. Soft clashes are clearance violations — the new line fits, but there is no room to weld it, no room to insulate it, no room for a valve handwheel to turn, or no maintenance access to the pump behind it. Soft clashes are what a legacy drawing cannot possibly reveal, because the drawing shows what was designed and not what accumulated around it.
The third class is temporal and it is the one that surprises people. Scaffolding, laydown, crane pick paths and rigging routes all have to fit through the same space, at the same time, in a specific order. Running access studies against the cloud is the same exercise as turnaround planning and scaffolding access work — the model shows not just whether the piece fits, but whether it can be carried to where it fits.
What a field interference actually costs
The cost of a dimensional error scales with how late it is found, and the scaling is not gentle. Found during design review against the cloud, a designer moves a line in an hour and the project absorbs it inside float. Found during a clash run before issue for construction, the design lead reroutes and reissues over a few days, still inside the engineering schedule. Found in the fabrication shop, the fabricator raises an RFI and re-cuts at shop rates, in shop conditions, off the critical path.
Found at field fit-up, everything changes. The correction happens at field labour rates, under permit, possibly at height, potentially requiring hot work on a live unit, with a crane committed and downstream activities queued behind it. A single occurrence costs a shift. Several occurrences compound, because each one displaces the next activity, and displacement travels straight through to startup. This is the row on the table where money is actually spent.
The Construction Industry Institute reports mean direct field rework above three percent of construction-phase cost on heavy industrial projects, from a database of more than 150 industrial construction projects (RS153-1), and its earlier Publication 10-2 estimated annual industrial construction rework losses at $15 billion. NIST's GCR 04-867 put the annual cost of inadequate interoperability across the US capital facilities industry at $15.8 billion on a 2002 basis. Dimensional uncertainty is a substantial, measured component of both.
Phasing a scan across a live operating unit
A live unit cannot be scanned in a single continuous campaign, and pretending otherwise produces a schedule that collapses in week two. The work phases by access condition rather than by discipline. Open, walkable areas go first, because permits are simple and escort demand is low. Congested rack levels, equipment enclosures and elevated platforms follow, scheduled against escort availability and operations tolerance rather than against the scan crew's convenience.
Three categories get deferred to the outage window on purpose. Anything requiring insulation removal, because a scanner returns cladding geometry and not pipe wall. Anything requiring confined space entry, which brings 29 CFR 1910.146 permit requirements, attendants, atmospheric monitoring and a rescue plan. And anything requiring elevated access beyond what a walkable platform provides. Attempting these during the live phase converts a passive survey into a permitted work activity with its own safety exposure.
The control network is established before any of it. Surveyed control points tied to the plant grid, carried through every registration, are what make phase four's data sit correctly against phase one's data six months later. Without that backbone, each phase is internally consistent and mutually misaligned, and the misalignment is discovered when someone extracts a tie-in dimension that spans two phases.
Specifying accuracy so the deliverable is contractable
Instrument specifications are the easy part and the least important. Leica publishes RTC360 3D point accuracy at 1.9 mm at 10 m, 2.9 mm at 20 m and 5.3 mm at 40 m, with range accuracy of 1.0 mm + 10 ppm, all at a 68 percent confidence level per JCGM 100:2008. Those are single-setup figures under the manufacturer's stated conditions. A unit as-built is hundreds of setups, and the accuracy that matters is what survives registration across all of them.
The USIBD Level of Accuracy Specification, at version 3.1 since January 2025, exists to make this contractable. It defines five bands, LOA10 through LOA50, derived from the five tolerance ranges in DIN 18710, and separates Measured Accuracy — the accuracy of the data captured — from Represented Accuracy, the accuracy of the models or drawings produced from that data. Published summaries place LOA20 at 15 mm to 5 cm and LOA30 at 5 mm to 15 mm, the two bands industrial as-built scopes are most often written against.
That separation is the useful part. A project can hold high measured accuracy and low represented accuracy at the same time, which is exactly what happens when a beautifully registered cloud is modelled quickly by someone with a deadline. Specify both, ask for registration residuals as a deliverable, and the vendor conversation moves from equipment brochures to engineering evidence in a single step.
Choosing the deliverable: cloud, model, or both
The cloud is evidence and it is complete. Every surface the scanner saw is in it, including the surfaces nobody thought to ask about, which is the whole reason scanning beats targeted survey. The model is interpretation and it is selective. Someone decided what to model, at what fidelity, and everything outside that decision is absent — not wrong, absent, which is harder to detect and easier to trust by mistake.
This drives a scoping rule that saves real money: model the systems the revamp touches plus the clash envelope around them, keep the cloud underneath for everything else. Modelling an entire unit to intelligent-object level produces a deliverable where most of the effort is never referenced by anyone. The conversion question — what gets modelled, at what level, in which format the design team can actually consume — is covered in detail on point cloud to CAD.
Format compatibility deserves an explicit line in the scope, not an assumption. A cloud delivered in a format the design team's platform cannot index is a cloud that gets converted at someone's cost, with the accuracy question reopened at the conversion step. Agree the exchange format, the coordinate system and the registration report before capture starts. Our 3D scanning services page sets out the capture and deliverable options across both.
When the as-built becomes something you maintain
A revamp as-built has a natural expiry date, and it is the day the next modification is made. Filed and forgotten, it becomes another archived drawing that was correct once — the exact failure mode it was bought to fix. Maintained, updated at each change and carried under change control, it becomes the dimensional base layer the plant works from, and the next revamp starts from measured reality rather than from another capture campaign.
This is the honest boundary between a scan deliverable and a digital twin. The difference is not visual fidelity and it is not the software. It is whether a named owner updates the model against a defined change-control process, and whether inspection and maintenance data attach to it. The comparison on digital twin versus IDMS covers where the geometric model ends and the inspection data management system begins, which is the question most owners are actually asking.
Where the model connects to work execution, the practical hook is the maintenance and project system — modification records, work orders and document control living in the same ERP that manages the rest of the plant, so the update trigger is a business process rather than someone's good intentions. Decide that ownership question before the scan, not after. Talk to us about scoping an as-built campaign against a specific revamp and a specific set of tie-ins.
Why are brownfield drawings wrong even when they exist?
Because modification outpaces documentation. Every outage moves supports, adds conduit, reroutes lines and installs temporary works that become permanent. Redlines are captured under time pressure, filed inconsistently and often never merged back into the master. Forty years of that produces an archive that is individually plausible and collectively wrong — and the errors concentrate exactly where revamp work happens, because that is where past work happened too.
How does scan data de-risk a tie-in?
By replacing an assumption with a measurement. A tie-in commits to a flange face position, a branch angle, a bolt hole orientation and a support elevation months before anyone can verify them. Taken from a registered cloud, those dimensions carry evidence and a stated accuracy. Taken from a legacy isometric, they carry a date and a hope. The spool either drops in or it does not, and that is decided here.
What does a clash discovered in the field actually cost?
It costs the compounding kind of time. A designer resolves an interference in an hour inside design float. A fabricator resolves it in weeks at shop rates, off the critical path. A field crew resolves it during an outage at field rates, on the critical path, with a crane committed and downstream activities queued behind. CII puts mean direct field rework above three percent of construction-phase cost on heavy industrial projects.
How is a scan phased across a live operating unit?
By area and by access condition rather than by discipline. Open, walkable areas go first while permits are simple. Congested rack levels and equipment enclosures follow, scheduled against escort availability. Anything requiring insulation removal, confined space entry under 29 CFR 1910.146 or elevated access is deferred to the outage window. The control network is established first so every phase registers to one coordinate system.
Point cloud or CAD model — which deliverable do we buy?
Both, in sequence, scoped by use. The registered cloud is evidence and it is complete; the model is interpretation and it is selective. Modelling everything is expensive and most of it is never referenced. Model the systems the revamp touches and the clash envelope around them, keep the cloud underneath for everything else, and specify the split explicitly in the scope.
How accurate should the as-built deliverable be?
Accurate enough to fabricate against, and stated in a specification rather than a sales sheet. PFI ES-3 sets linear fabricating tolerance at ±1/8 in for pipe 10 in and under. The USIBD Level of Accuracy Specification gives contract language for this, separating Measured Accuracy — the captured data — from Represented Accuracy, the models or drawings derived from it. Specify both.