From point cloud to a clean Revit model — best practices for industrial as-builtFrom point cloud to a clean Revit model — best practices for industrial as-builtFrom point cloud to a clean Revit model — best practices for industrial as-built

A point cloud from a terrestrial laser scanner is raw data. What the plant engineer actually needs at the end is a model — RVT for the architects, DWG for the design office, IFC for the general contractor, occasionally STEP for mechanical engineering. The leap between them is the part of the project that regularly takes longer than planned.

Why "modelling from the point cloud" takes longer than expected. Three recurring sources of error.

• Over-modelling. A scan resolves every weld seam and every rivet line on a steel beam. The BIM model needs none of it. If you digitise everything anyway, you miss the delivery date by weeks.
• No tolerance agreement. "How accurate should the model be?" gets settled at the start of a project or at the end — rarely in the middle. Without a documented tolerance every element gets challenged in review.
• Registration drift in large plants. Across 50+ scan positions, small target errors add up to several centimetres. Without a control-point network on the site, that only becomes visible during modelling — when pipes suddenly do not line up with vessel nozzles.

The Sachtleben workflow in five steps. Across multiple industrial as-built projects — most recently at a Refratechnik plant and an Aurubis site — this sequence has proven itself:

1. Scan. Riegl VZ-400i or VZ-2000i depending on range. Setup spacing 15–25 m indoors, 30–50 m outdoors. Targets every 8–12 setups as registration anchors. Resolution: 6 mm@10 m for plant detail, 20 mm@10 m for pure terrain survey.

2. Register. Point-cloud registration in RiSCAN PRO or Cyclone REGISTER 360. Target: per-setup residual under 5 mm, full network under 10 mm. If you do not hit that, go back to the scan — do not push forward.

3. Segment. Split the cloud into functional zones (steel, piping, vessels, building envelope). Each zone gets its own modelling discipline in Revit or AutoCAD Plant 3D. Classification is mostly semi-automatic via CloudCompare or Faro Webshare — manual cleanup remains.

4. Model. Modelling in Revit (RVT) or AutoCAD/Plant 3D (DWG). This is where the tolerance agreement matters: 5–10 mm geometric tolerance for components that are fabrication-relevant; 20–50 mm for pure as-built documentation. That gets agreed in writing up front, including the list of geometries that are not modelled (cable trays under 50 mm, bolted connections, insulation under 20 mm, etc.).

5. QA. Render the model back into the point cloud (cloud-vs-model comparison in Faro Webshare or Recap). Deviations above tolerance are colour-coded. The report ships with the model — transparent, not hidden.

Realistic tolerances per geometry type. What we commit to in industrial projects, based on what plant engineers and architects actually need:

Geometry element	Realistic tolerance	Note
Steel beams (profile + position)	±5 mm	Profile from catalog, position from scan
Pipes DN ≥ 50	±10 mm	Centerline; nozzles modelled separately
Wall surfaces (masonry/concrete)	±10 mm	Modelled as flat surface
Foundations (visible edges)	±10 mm	Underground portions not modelled
Stairs + railings	±15 mm	Standard families from library
Platforms + grating	±10 mm	Structure yes, decking as surface only
Columns (round/rectangular)	±5 mm	Profile + axis position
Equipment outlines (BIM-box)	±20 mm	Vessels as bounding box, not detail

Delivery formats. Standard today: RVT (Revit native), DWG (3D in AutoCAD/Plant 3D), IFC 4 for open exchange between disciplines, STEP only on request for mechanical-engineering components. We hand over the point cloud itself in parallel as RCP or E57 — so the recipient can do their own detail measurements later without re-scanning.

One recommendation. The most common project killer is neither the scan nor the modelling — it is a missing tolerance + delivery spec at the start. Whoever invests 30 minutes in a written list of "these geometries will be modelled, at this tolerance, in this format" saves 30 hours in the review.

More at /scan-to-cad/, /bulk-inventory/ and in the FAQ.

A point cloud from a terrestrial laser scanner is raw data. What the plant engineer actually needs at the end is a model — RVT for the architects, DWG for the design office, IFC for the general contractor, occasionally STEP for mechanical engineering. The leap between them is the part of the project that regularly takes longer than planned.

Why "modelling from the point cloud" takes longer than expected. Three recurring sources of error.

• Over-modelling. A scan resolves every weld seam and every rivet line on a steel beam. The BIM model needs none of it. If you digitise everything anyway, you miss the delivery date by weeks.
• No tolerance agreement. "How accurate should the model be?" gets settled at the start of a project or at the end — rarely in the middle. Without a documented tolerance every element gets challenged in review.
• Registration drift in large plants. Across 50+ scan positions, small target errors add up to several centimetres. Without a control-point network on the site, that only becomes visible during modelling — when pipes suddenly do not line up with vessel nozzles.

The Sachtleben workflow in five steps. Across multiple industrial as-built projects — most recently at a Refratechnik plant and an Aurubis site — this sequence has proven itself:

1. Scan. Riegl VZ-400i or VZ-2000i depending on range. Setup spacing 15–25 m indoors, 30–50 m outdoors. Targets every 8–12 setups as registration anchors. Resolution: 6 mm@10 m for plant detail, 20 mm@10 m for pure terrain survey.

2. Register. Point-cloud registration in RiSCAN PRO or Cyclone REGISTER 360. Target: per-setup residual under 5 mm, full network under 10 mm. If you do not hit that, go back to the scan — do not push forward.

3. Segment. Split the cloud into functional zones (steel, piping, vessels, building envelope). Each zone gets its own modelling discipline in Revit or AutoCAD Plant 3D. Classification is mostly semi-automatic via CloudCompare or Faro Webshare — manual cleanup remains.

4. Model. Modelling in Revit (RVT) or AutoCAD/Plant 3D (DWG). This is where the tolerance agreement matters: 5–10 mm geometric tolerance for components that are fabrication-relevant; 20–50 mm for pure as-built documentation. That gets agreed in writing up front, including the list of geometries that are not modelled (cable trays under 50 mm, bolted connections, insulation under 20 mm, etc.).

5. QA. Render the model back into the point cloud (cloud-vs-model comparison in Faro Webshare or Recap). Deviations above tolerance are colour-coded. The report ships with the model — transparent, not hidden.

Realistic tolerances per geometry type. What we commit to in industrial projects, based on what plant engineers and architects actually need:

Geometry element	Realistic tolerance	Note
Steel beams (profile + position)	±5 mm	Profile from catalog, position from scan
Pipes DN ≥ 50	±10 mm	Centerline; nozzles modelled separately
Wall surfaces (masonry/concrete)	±10 mm	Modelled as flat surface
Foundations (visible edges)	±10 mm	Underground portions not modelled
Stairs + railings	±15 mm	Standard families from library
Platforms + grating	±10 mm	Structure yes, decking as surface only
Columns (round/rectangular)	±5 mm	Profile + axis position
Equipment outlines (BIM-box)	±20 mm	Vessels as bounding box, not detail

Delivery formats. Standard today: RVT (Revit native), DWG (3D in AutoCAD/Plant 3D), IFC 4 for open exchange between disciplines, STEP only on request for mechanical-engineering components. We hand over the point cloud itself in parallel as RCP or E57 — so the recipient can do their own detail measurements later without re-scanning.

One recommendation. The most common project killer is neither the scan nor the modelling — it is a missing tolerance + delivery spec at the start. Whoever invests 30 minutes in a written list of "these geometries will be modelled, at this tolerance, in this format" saves 30 hours in the review.

More at /scan-to-cad/, /bulk-inventory/ and in the FAQ.

A point cloud from a terrestrial laser scanner is raw data. What the plant engineer actually needs at the end is a model — RVT for the architects, DWG for the design office, IFC for the general contractor, occasionally STEP for mechanical engineering. The leap between them is the part of the project that regularly takes longer than planned.

Why "modelling from the point cloud" takes longer than expected. Three recurring sources of error.

• Over-modelling. A scan resolves every weld seam and every rivet line on a steel beam. The BIM model needs none of it. If you digitise everything anyway, you miss the delivery date by weeks.
• No tolerance agreement. "How accurate should the model be?" gets settled at the start of a project or at the end — rarely in the middle. Without a documented tolerance every element gets challenged in review.
• Registration drift in large plants. Across 50+ scan positions, small target errors add up to several centimetres. Without a control-point network on the site, that only becomes visible during modelling — when pipes suddenly do not line up with vessel nozzles.

The Sachtleben workflow in five steps. Across multiple industrial as-built projects — most recently at a Refratechnik plant and an Aurubis site — this sequence has proven itself:

1. Scan. Riegl VZ-400i or VZ-2000i depending on range. Setup spacing 15–25 m indoors, 30–50 m outdoors. Targets every 8–12 setups as registration anchors. Resolution: 6 mm@10 m for plant detail, 20 mm@10 m for pure terrain survey.

2. Register. Point-cloud registration in RiSCAN PRO or Cyclone REGISTER 360. Target: per-setup residual under 5 mm, full network under 10 mm. If you do not hit that, go back to the scan — do not push forward.

3. Segment. Split the cloud into functional zones (steel, piping, vessels, building envelope). Each zone gets its own modelling discipline in Revit or AutoCAD Plant 3D. Classification is mostly semi-automatic via CloudCompare or Faro Webshare — manual cleanup remains.

4. Model. Modelling in Revit (RVT) or AutoCAD/Plant 3D (DWG). This is where the tolerance agreement matters: 5–10 mm geometric tolerance for components that are fabrication-relevant; 20–50 mm for pure as-built documentation. That gets agreed in writing up front, including the list of geometries that are not modelled (cable trays under 50 mm, bolted connections, insulation under 20 mm, etc.).

5. QA. Render the model back into the point cloud (cloud-vs-model comparison in Faro Webshare or Recap). Deviations above tolerance are colour-coded. The report ships with the model — transparent, not hidden.

Realistic tolerances per geometry type. What we commit to in industrial projects, based on what plant engineers and architects actually need:

Geometry element	Realistic tolerance	Note
Steel beams (profile + position)	±5 mm	Profile from catalog, position from scan
Pipes DN ≥ 50	±10 mm	Centerline; nozzles modelled separately
Wall surfaces (masonry/concrete)	±10 mm	Modelled as flat surface
Foundations (visible edges)	±10 mm	Underground portions not modelled
Stairs + railings	±15 mm	Standard families from library
Platforms + grating	±10 mm	Structure yes, decking as surface only
Columns (round/rectangular)	±5 mm	Profile + axis position
Equipment outlines (BIM-box)	±20 mm	Vessels as bounding box, not detail

Delivery formats. Standard today: RVT (Revit native), DWG (3D in AutoCAD/Plant 3D), IFC 4 for open exchange between disciplines, STEP only on request for mechanical-engineering components. We hand over the point cloud itself in parallel as RCP or E57 — so the recipient can do their own detail measurements later without re-scanning.

One recommendation. The most common project killer is neither the scan nor the modelling — it is a missing tolerance + delivery spec at the start. Whoever invests 30 minutes in a written list of "these geometries will be modelled, at this tolerance, in this format" saves 30 hours in the review.

More at /scan-to-cad/, /bulk-inventory/ and in the FAQ.