LiDAR and laser scanning for US refinery turnaround planning and as-built capture

Two scanning windows, and confusing them wastes the money

A turnaround has two distinct scan windows serving two different customers. The pre-turnaround scan happens while the unit still runs, six to twelve months before the shutdown date, and it exists to serve planning: tie-in locations, spool dimensions, scaffold envelopes, crane pick paths, laydown geometry. The execution scan happens after demolition and during reinstatement, and it exists to serve the record: what was actually installed, where the new nozzle actually landed, which line was rerouted around an obstruction that appeared on no drawing. Owners who budget for only one usually buy the wrong one.

The pre-turnaround scan pays inside this cycle. Every dimension it settles is a dimension a crew does not chase at two in the morning with a tape measure and a radio while a superintendent watches the clock. The execution scan pays in the next cycle and the one after, because it is the only moment in the asset's life when insulation is off, scaffolding is up and internals are exposed. Skip it and the next planner inherits the same wrong drawings you inherited, and buys the same pre-turnaround scan all over again.

Both feed the same downstream artefact. Once the unit is back online, that registered cloud is the geometric base layer under turnaround planning and scaffolding access work for the next event, and the surface every future thickness reading gets pinned to. A scan bought as a one-off measurement exercise is a scan bought at full price and used at a fraction of its value.

What the scan actually replaces, line by line

Three things go away. The first is manual field measurement — the two-person crew with a tape, a laser distance meter and a clipboard, working under permit, on scaffold, in an operating unit, producing dimensions that are correct only for the questions someone thought to ask that morning. The second is the repeat site visit, which is the expensive one: every dimension nobody thought to take becomes a new permit, a new escort, a new travel day and a new week on the engineering schedule. The third is field clash discovery.

Field clash discovery is the failure mode that hurts. A designer routes a new 6-inch line through what the P&ID and the old model show as clear space. The steel is fabricated, shipped and rigged into place during the outage, and then it does not fit, because a conduit run was added in 2011 and a pipe support was moved in 2018 and neither made it back to the drawing. That discovery lands on the critical path, at field labour rates, with a crane already committed.

The Construction Industry Institute puts mean direct field rework above three percent of construction-phase cost on heavy industrial projects, drawn from a database of more than 150 industrial construction projects (RS153-1). NIST separately estimated 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). Dimensional error is a large, measured share of both figures.

Tie-ins and spool fabrication straight from the point cloud

The tie-in is where scan data earns its keep, because a tie-in is a dimensional promise made months before anyone can verify it. Cutting into an existing header means committing to a flange face position, a branch angle, a bolt hole orientation and a support elevation that all have to be right when the line is dead and the clock is running. Take those dimensions off a registered cloud and they come with evidence attached; take them off a 1978 isometric and they come with hope attached.

PFI ES-3 sets the standard fabricating tolerance band: ±1/8 in (3.0 mm) on linear dimensions for pipe 10 in and under, ±3/16 in (5.0 mm) for 12 through 24 in, and ±1/4 in (6.0 mm) over 24 in through 36 in. A spool fabricated to that tolerance against a wrong datum is still a wrong spool. Scanning moves the argument from tolerance to datum, which is the argument that actually decides whether the piece drops in.

Practically, the fabricator receives extracted geometry rather than the raw cloud — dimensioned isometrics, a modelled solid, or both, tied to plant coordinates. The field-fit weld strategy changes with it. Instead of leaving generous field-fit allowances on every spool and grinding to suit, the fabricator ships to final dimension on the runs that were measured, and reserves field-fit only where the cloud was genuinely occluded. That is where shift hours come back.

The schedule argument, stated plainly

AP-Networks, working from a turnaround database of more than 1,350 observations across onshore and offshore upstream, gas processing, refining, chemicals and power generation, reports that more than two-thirds of turnarounds exceed planned cost and schedule by ten percent or suffer a trip after startup, and that forty percent overrun cost or schedule by more than thirty percent. That is the industry baseline any scanning business case has to be measured against.

Scanning does not touch most of that. It does not accelerate a catalyst change, shorten a hydrotest or make a stress-relief cycle finish sooner. What it addresses is the subset of overrun driven by dimensional discovery — the hours lost between a piece not fitting and the piece being made to fit. Those hours are concentrated at fit-up, which sits late in the sequence, which means they compound: a shift lost at fit-up pushes reinstatement, which pushes tightness testing, which pushes startup.

This is why the honest framing is narrow. Scanning buys back the days that measurement error takes, and nothing else. That framing survives contact with a turnaround manager who has seen every technology pitch there is, and it is the framing that gets the scan into the pre-turnaround budget rather than the innovation budget.

Scanning a live, hot, running unit

Terrestrial laser scanning on an operating unit is a passive optical measurement. There is no contact with the asset, no hot work, no line break and no interruption to process. What governs it is the site's permit-to-work system and area classification, which control what electronic equipment enters a classified area and under what conditions — the same controls that govern a camera, a tablet or a gas detector. Escort availability and route planning constrain the schedule far more than the instrument does.

Occlusion is the real technical limit. A scanner measures what it can see, and an operating unit is full of things it cannot: insulated lines that hide the pipe wall, congested rack levels where the upper tiers shadow the lower, equipment behind permanent guarding. The answer is setup density and setup geometry rather than a better instrument. Planning where the scanner stands is the skilled part of the job, and it is decided by walking the unit first.

Control is the other half. A cloud that is internally consistent but floating relative to plant coordinates is useless for a tie-in, because the fabricator needs to know where the flange face is in the plant, not where it is relative to setup seventeen. Surveyed control points, tied to the site grid and carried through registration, are what make the cloud dimensionally trustworthy across a whole unit rather than within a single room.

Capturing modifications during execution, while the metal is bare

The execution window is short and it never comes back. During the shutdown, insulation is stripped, scaffolding is standing, vessels are open, exchanger bundles are pulled and foundations are exposed. Every one of those conditions costs real money to create, and all of them are already paid for. Capturing geometry at that moment is the cheapest dimensional data the asset will ever produce, and it is the only data that shows bare pipe wall rather than cladding.

What to capture is a scope decision made in advance, not on the day. Nozzle positions and flange faces on replaced equipment, the actual routing of any line that moved, new support steel, exposed foundation dimensions, and the interior geometry of any vessel that will be entered again. Crews executing the turnaround are not the crews doing the scanning; the scan team works the gaps between activities, which means the schedule has to make room for them before the freeze.

Data turnaround matters here in a way it does not pre-turnaround. If a scan of a demolished section is registered and issued within the shutdown, the design change riding on it can be resolved while the unit is still down. If it is issued four weeks later, it becomes a documentation exercise. The procedure and workflow definition for who scans what, when, and against which deliverable clock is the difference between the two outcomes.

Accuracy you can put in a contract

Instrument accuracy is published and verifiable. Leica Geosystems states RTC360 3D point accuracy of 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 and angular accuracy of 18 in, all at a 68 percent confidence level per JCGM 100:2008. Those are single-setup figures under the manufacturer's stated conditions, and they are the floor, not the delivered accuracy of a multi-setup registered project.

Delivered accuracy is a registration question. Error accumulates across setups, and a unit scan is hundreds of setups. This is why the survey control network, the registration method and the residual reporting are the specifications worth arguing about — not the scanner model. Ask for registration residuals as a deliverable and the conversation moves from marketing to engineering in one step.

The USIBD Level of Accuracy Specification, at version 3.1 since January 2025, gives buyers language for this. It defines five bands, LOA10 through LOA50, inspired by the five tolerance ranges in DIN 18710, and — critically — separates Measured Accuracy, the accuracy of the captured data, from Represented Accuracy, the accuracy of the models or drawings derived from it. Published summaries place LOA20 at 15 mm to 5 cm and LOA30 at 5 mm to 15 mm, the bands most industrial as-built work is written against.

Where turnaround scanning concentrates in the United States

The EIA counted 130 operable refineries with 18.2 million b/cd of operable atmospheric distillation capacity on 1 January 2026, down about 250,000 b/cd from a year earlier. The Gulf Coast, PADD 3, holds more than half of that capacity. The named clusters are Port Arthur and Beaumont, the Houston Ship Channel and Baytown, Texas City, Corpus Christi, Lake Charles, and the Mississippi River corridor from Baton Rouge to Garyville and Norco.

Scale concentrates the exposure. Motiva Port Arthur is the largest US refinery at 656,000 b/cd; Marathon Galveston Bay is the largest on a stream-day basis at 678,000 b/sd; Marathon Garyville runs 617,000 b/cd and ExxonMobil Beaumont 612,000 b/cd, all as of 1 January 2026. A turnaround on any of these is a project of a size where dimensional discovery is not a nuisance, it is a line item.

Outside the Gulf, the work sits in the Midwest at Whiting, Wood River, Toledo and Catlettsburg; in the Rockies around Salt Lake City and Billings; on the West Coast in the Los Angeles basin, the Bay Area and Anacortes; and on the East Coast at Delaware City and Marcus Hook. Two 2025 closures — LyondellBasell Houston at 263,776 b/cd and Phillips 66 Los Angeles at 138,700 b/cd — removed roughly 400,000 b/d and put more schedule pressure on the units still running.

What the point cloud becomes after the unit restarts

A point cloud that is archived after the turnaround is a photograph. A point cloud that is maintained, updated at each modification and carried forward as the geometric base layer becomes something else: the surface that inspection data attaches to. That is the actual transition from scan data to a digital twin, and it has nothing to do with rendering quality and everything to do with whether anyone owns the update cycle.

The immediate payoff is spatial inspection data. Once corrosion monitoring locations exist as coordinates on real geometry rather than as rows in a spreadsheet, thickness trending on asset geometry shows where wall loss clusters on the asset rather than where it clusters in a table — which is how a thinning band on the outlet side of an elbow becomes visible as a pattern instead of five unrelated readings.

The decision worth making before the scan, not after, is who owns that model afterwards and against what change-control process. If the answer is nobody, buy the scan for this turnaround, use it for this turnaround, and price it accordingly. If the answer is a named owner, specify the deliverable so it survives — which is a conversation worth having on the way in. Our 3D scanning services page covers the capture side, and contact us for a scope and quote on a specific unit.

Request a consultation