Reverse Engineering Cast Parts

A sand-cast valve body on our scan table looks like a simple gray lump. It is actually a record of a dozen foundry decisions encoded in metal - taper angles, fillet radii, parting lines, and shrinkage compensation - none of which appear labeled on the surface. Copy the geometry without decoding those decisions and the first pour produces scrap.

This article covers our full workflow for reverse engineering services on castings: scanner selection, point cloud processing, draft angle extraction, shrink back-calculation, and foundry-ready delivery.

---

Why Cast Parts Are the Hardest Geometry to Reverse Engineer

Three problems stack on top of each other, and machinists or product designers who have not worked castings before consistently underestimate all three.

As-cast surface noise. Green sand casting produces Ra 12.5-25+ µm surface roughness - that is micrometers, not millimeters. A handheld scanner reading 0.040 mm point accuracy is picking up genuine surface data, but that surface is inherently rough. Naively meshing the raw cloud gives you a lumpy solid with no clean geometric features.

Applied draft on every vertical wall. Every wall that pulls out of the mold has taper - typically 1°-3° per side for green sand, less for die casting. That taper is real geometry that must be preserved in the CAD model. It is not an error to be averaged away.

Shrinkage already baked in. The physical part on your scan table has already shrunk. If you copy its geometry exactly into CAD and send it to a pattern shop, the new castings will be undersized by the shrink amount on every dimension. Copying scan geometry without applying shrink allowance will produce undersized castings on every dimension - a costly outcome on any production run.

Common orphaned-casting scenarios: valve bodies where the OEM foundry closed, pump housings where the original manufacturer was acquired and pattern inventory was destroyed, gear cases for industrial machinery discontinued in the 1990s, and custom brackets where the person who made them retired and took the drawings. All share the same profile - a serviceable physical part, no pattern, no drawings, and a production need.

Contrast these with machined parts. A turned shaft has clean cylindrical datums, flat shoulders, and a surface finish of Ra 0.8-3.2 μm. You can lock it in a V-block and reference from it confidently. A sand-cast housing has Ra 12.5-25+ µm surface roughness, variable wall thickness from core shift, and no reliably flat surface anywhere. The discipline is fundamentally different.

---

Scanning Gear Selection: Structured-Light Arm vs. TLS for Cast Parts

The right tool depends on part size and fillet detail required.

Arm scanner vs. handheld scanner - when each wins. A contact-and-non-contact arm scanner is a hybrid: switch between the laser line probe and a hard probe tip on the same device. That makes it ideal when bore IDs need to be captured at touch-probe accuracy (±0.025 mm) on the same setup as the optical surface scan. A photogrammetry-tracked handheld scanner such as the Creaform MetraSCAN is fully portable - no fixed base, no worktable required. On a 500 mm pump casing that cannot be moved, we bring the handheld scanner to the part. On a precision bore that needs sub-0.030 mm capture, an arm scanner is the appropriate industry choice. The two tools are not interchangeable; choosing the wrong one wastes a re-scan.

For small-to-medium castings - the 250 mm valve body, the 400 mm pump casing - a handheld scanner such as the Creaform MetraSCAN (the BLACK+ variant carries 0.035 mm accuracy and 0.025 mm resolution; the BLACK+ Elite carries 0.025 mm accuracy), or an arm scanner where sub-0.030 mm bore capture is required, provides the point accuracy needed to recover R3-R10 mm fillets and to detect the 0.1-0.5 mm parting-line flash ridge that tells us where the mold parted.

For large castings, we run our terrestrial laser scanner (Trimble X7) in close-range mode at 0.3-1.0 m standoff. That gives 0.5-1.0 mm point accuracy - adequate for draft angles and gross geometry, with a supplemental handheld scan on critical fillet zones.

Surface prep is not optional. Gray iron and aluminum castings are specular enough to cause scan noise or dropout. We spray with Magnaflux Spotcheck SKD-S2 developer - a matte white, alcohol-carrier suspension that dries in 60-90 seconds to a 0.02-0.05 mm coat. That coat thickness is thin enough that it does not measurably affect R6 mm or larger fillet geometry (deviation < 0.03 mm on an R6 surface), but we do note it in the project report and clean it off with acetone before any fillet measurements below R4 mm where the coat thickness becomes a meaningful fraction of the radius. On parts where coating is not allowed, we adjust scan angle to minimize specular return - but coating is always the cleaner path.

A typical 250 mm valve body requires 12-18 scan positions to achieve full coverage including flange faces, boss features, and accessible internal bores. Output is 8-15 million points at roughly 600 MB in .e57 format before registration.

---

Step 1 - Point Cloud Processing: Filtering As-Cast Surface Noise

Sand casting’s Ra 12.5-25+ µm surface roughness shows up in the scan as high-frequency noise that is indistinguishable from geometry at the raw cloud level. The objective is to attenuate that noise without destroying the fillet radii and draft angle data we came to recover.

Our workflow in Autodesk ReCap Pro: import the .e57 from each scan position, run outlier removal with a deviation filter set at 2σ (removes floating points and scan shadow artifacts), then apply a unified distance filter at 0.15 mm for small castings under 300 mm. For castings in the 600 mm-1 m range, we open that filter to 0.25-0.35 mm. The goal is suppressing texture noise while preserving local geometry.

Die castings and investment castings are a different story. Their surface finish is Ra 1.6-3.2 μm - orders of magnitude smoother than sand casting. We apply minimal filtering and preserve fillet geometry at full resolution because the noise floor is low enough that aggressive filtering would remove real geometry.

Registration accuracy is the gate before moving downstream:

• Handheld/arm scans: max deviation ≤ 0.05 mm across all registered positions
• TLS on large housings: max deviation ≤ 0.3 mm

We verify with a color-deviation map in ReCap before exporting. If any registration pair shows deviation above threshold, we rescan that position rather than average it away.

Output at this stage: a registered, cleaned .e57 or .rcp point cloud, typically 2-5 GB for a medium casting, ready for mesh reconstruction.

---

Step 2 - Mesh and Solid Model: Recovering Draft Angles from Noisy Walls

This is where casting-specific expertise matters most.

We mesh in Geomagic Design X using Accurate mode rather than Smooth. Smooth mode averages wall angles - which destroys draft information. Accurate mode preserves the wall angle at the cost of surface appearance, and that is the correct trade-off for foundry pattern work.

Draft angle measurement workflow in Geomagic Design X. We use the Region Group command (Accuracy Analyzer panel) to auto-segment wall surfaces from the mesh by curvature threshold, typically set at 3°-5° to separate nominally planar wall zones from fillet transitions. Each segmented region is promoted to a Best-Fit Plane using the Insert → Plane → Best-Fit function, fitting only the interior of the region and excluding the 2 mm boundary band adjacent to fillets where curvature blending can bias the plane normal. We then measure the included angle between that best-fit plane and the designated parting plane using the 3D Compare → Angular Deviation tool. The result is logged as a named annotation in the Design X tree - for example, “Core Wall A: 1.7° draft” - so the pattern shop can reference the annotated model alongside the STEP file rather than reading a separate spreadsheet.

Typical results by process:

The critical point: the scanned part shows the draft angle as it exists in the finished casting. To make a new pattern, we reproduce that same angle - we do not clean up the taper or round it to the nearest integer degree. The foundry needs what actually worked the first time.

Fillet recovery. In Geomagic Design X we use the Region Growing Surface Fit tool to segment fillet zones from adjacent wall surfaces. The algorithm fits an analytical radius to the fillet band, producing a clean R-value - R6.0 mm, R8.0 mm - rather than a faceted mesh arc. This matters for CAM: a toolpath generated from a faceted mesh arc produces a rougher pattern finish than a path generated from an analytical radius.

Parting line detection - distinguishing flash from geometry. The scan almost always shows a subtle ridge 0.1-0.5 mm proud of the nominal wall surface - residual metal that squeezed into the mold parting plane. In Geomagic Design X, we identify this ridge by running Curvature Analysis on the mesh: flash presents as a sharp curvature discontinuity (local curvature > 40°/mm) against the broad, low-curvature wall surface on either side. Legitimate blending radii show a smooth, progressive curvature gradient. Once identified, we flag the flash ridge as a candidate parting plane locator and confirm it manually by checking that the ridge runs continuously around the part perimeter - a random ding or repair mark will not close the loop, but a true parting line will. Borderline cases (< 0.15 mm ridge height on a rough-surface casting) get a manual review note in the project file with a screenshot of the curvature map so the foundry can make the final call.

Deliverable at this stage: a watertight STEP or IGES solid with annotated draft angles, parting line geometry, and a fillet radius table listing every R-value in the model.

For a complete walkthrough of how we take point cloud data through this pipeline, see our post on turning point cloud data into usable CAD output.

---

Step 3 - Shrinkage Back-Calculation: Making the Pattern Bigger Than the Part

Every casting process shrinks during solidification. The part on our scan table has already shrunk. If we copy its geometry exactly into a STEP file and hand it to a pattern shop, the new castings will be undersized by the shrink amount on every dimension.

Shrink allowance by material:

For a 300 mm gray iron housing, a 1.0% shrink allowance adds 3.0 mm to the longest dimension. On a bore ID that feeds a shaft fit, that is the difference between a sliding fit and an interference fit.

The math is applied by scaling the verified CAD solid in the modeling environment. For most parts we apply a uniform scale factor. But for complex geometries with significant section thickness variation - a thick-walled flange next to a thin rib - thick sections cool slower and shrink more. We flag those zones and, where the foundry has simulation data from MAGMASOFT or ProCAST, apply local oversize values rather than a blanket scale.

What local oversize looks like numerically. For parts with significant section thickness variation, MAGMASOFT or ProCAST simulation can identify zones where local shrink exceeds the global rate by 0.1-0.2%. In a gray iron part where a nominal 1.0% global shrink is applied, thick-section zones such as a heavy-wall flange or a dense gusset cluster may carry 1.15-1.25% local shrink - translating to 0.7-1.1 mm of additional oversize correction on those specific features versus the global scale. We apply those local corrections as direct dimension overrides in the STEP model rather than additional global scaling, and document each override in the title block so the pattern shop knows exactly which features carry non-standard shrink. Applying only a global scale without local correction on heavy-walled sections is a common cause of undersized bores - a bore rework outcome that no production run budget plans for.

We document the applied shrink factor explicitly in the drawing title block and embed it in the STEP file metadata. A STEP file with undocumented shrink is an incomplete deliverable.

---

Step 4 - Delivering a Foundry-Ready Pattern Package

A complete foundry pattern package includes:

1. STEP AP214 solid with shrink applied, draft angles annotated, and parting line flagged
2. 2D drawing (PDF + DXF) with GD&T callouts on all machined surfaces, parting line noted, material and shrink factor in the title block
3. Color deviation report comparing the final CAD model to the registered scan - target: ≤ 0.1 mm 2σ on functional surfaces
4. Fillet/radius table listing every R-value in the model with its location reference
5. Shrink factor confirmation memo - a one-page document stating material, casting process, shrink applied, and the foundry contact who validated it

The STEP file feeds directly into CAM. Pattern shops typically run Mastercam or Fusion 360, and STEP AP214 is readable by both without translation loss. Typical pattern materials are RenShape tooling board (fast to machine, good for prototype patterns), aluminum (preferred for production patterns over 500 pieces), and hardwood composite for short-run or budget work.

We recommend a first-article inspection scan on every pattern project: once the foundry pours the first test casting, we re-scan it and overlay it against the original part scan. Deviations greater than 0.5 mm on machined pads indicate a pattern correction or process adjustment is needed before production commitment.

Timelines and cost:

---

Injection Mold Cavity Reverse Engineering: Where It Differs from Foundry Work

Injection-molded plastic parts share the draft-and-fillet challenge but layer on three more complications: gate vestige geometry, sink marks over thick sections, and warpage - all of which distort the true nominal geometry. A single scan of one warped part produces a warped CAD model, which is the single most common error seen when mold cavity work is attempted from a single-sample scan.

Warpage and multi-sample averaging. Scanning multiple samples from the same production run and averaging them in CloudCompare reduces warpage bias significantly. On a thin-walled lid part, a single scan may show 0.3-0.5 mm of apparent bow; averaging five to ten parts can suppress that to 0.05-0.15 mm - a meaningful difference when specifying rib depth for a steel mold. For example, on a flat-panel lid geometry at roughly 150 mm span, the recovered long-axis cavity dimension is derived from the desired part dimension divided by (1 - shrink rate), and cavity depth on stack ribs - the feature most prone to per-part variation - can read 0.15-0.20 mm deeper on a single-part scan than on the averaged multi-part result. That difference matters: a mold cut too deep on a rib requires a steel weld repair to correct, a $400-$900 cost on a simple rib.

Warpage suppression protocol: scan 5-10 samples from the same production run, import all meshes into CloudCompare, align all to one reference using ICP registration, then compute the average mesh. Warpage bias that can run 0.3-1.0 mm on thin-walled parts (< 2 mm wall) is suppressed to 0.05-0.15 mm in the averaged result. The averaged geometry is a substantially better representation of intended nominal shape than any single part.

Draft angles for injection molds:

Shrink math for mold cavities - direction reverses from casting. For castings, we scale the pattern larger because the metal shrinks. For mold cavities, the cavity must be larger than the desired finished part because the injected resin shrinks as it cools. The formula is: cavity dimension = desired part dimension ÷ (1 - shrink rate).

Glass-filled resins require special attention. GF nylon and GF PP shrink differently along the flow direction versus transverse to it - anisotropic shrink of up to 0.5% differential. On a part with a clear gate location (identifiable gate vestige on the surface), we note the flow axis in the deliverable so the mold shop can apply directional shrink corrections if their simulation supports it. On parts where the gate vestige is fully trimmed and flow direction is ambiguous, we note that limitation explicitly and recommend the mold shop run Moldflow or Autodesk Simulation Moldflow on the recovered cavity geometry before cutting steel.

The deliverable for a mold shop is a STEP cavity solid scaled for shrink, a defined parting surface, side-action pull directions, and runner/gate geometry where the vestige is recoverable. Understanding how reverse engineering differs from product design scan work is important context - this workflow recovers the cavity as-was, not a redesigned part.

---

Sample Workflow: Reverse Engineering a Gray Iron Valve Body with No Original Pattern

The following describes how we approach a part of this type. For a 6-inch gate valve body in gray iron where the original pattern no longer exists - no drawings, no 3D files, no pattern - the workflow proceeds as follows.

Scan approach: A handheld scanner such as the Creaform MetraSCAN at 0.040 mm single-point accuracy is appropriate for this geometry. The part is mounted stably for scanning, and approximately 16 scan positions are typically required for full coverage. Surface prep with Magnaflux Spotcheck SKD-S2 developer eliminates scan dropout on bore IDs and flange faces on the moderately specular gray iron surface. Total scan time: approximately 4 hours including setup and position verification.

Processing: ReCap Pro registration across all scan positions; target max registration deviation 0.038 mm, inside the 0.05 mm threshold. The cleaned .rcp cloud is imported into Geomagic Design X. Running discrete draft angle measurements across all vertical wall zones using the Best-Fit Plane workflow described above typically confirms 1.5°-2.0° draft consistent with green sand process. The parting line is identified from a flash ridge on the outer barrel - confirmed as a true parting line rather than a repair mark by the continuous closed-loop curvature break running the full circumference. A 0.10-0.25 mm flash ridge is typical on sand-cast iron; borderline cases below 0.15 mm receive a manual review note and a curvature-map screenshot so the foundry can make the final call.

Zone-by-zone deviation targets for the final CAD model vs. scan:

The 0.18 mm maximum deviation on the barrel OD is on an as-cast, non-machined wall - acceptable for a sand-cast part. The machined pads (flange faces and bore ID) target 0.07-0.09 mm, well inside the ≤ 0.1 mm target for functional surfaces.

Shrink applied: 1.0% uniform scale for gray iron. Bore IDs - the functional sealing surfaces - are overridden to the client’s nominal valve seat dimension where the engineering team has seat tolerance requirements that supersede as-cast geometry.

Delivery: STEP AP214 + 2D drawing with GD&T on all machined pads + color deviation report + shrink confirmation memo. A CNC pattern shop can machine an aluminum pattern from the STEP file in approximately 6 business days.

---

When 3D Scanning Alone Is Not Enough: Internal Features & CT Scanning

Optical and laser scanning are line-of-sight technologies. For cast parts with blind internal passages - oil galleries, coolant jackets, cored valve chambers, hydraulic manifold passages - the inner geometry is completely invisible to surface scanning.

Options for recovering internal features:

• Destructive sectioning: Cut the part at key cross-sections, scan each face, reconstruct internal geometry from section profiles. Useful when only one sample is available and the sample can be consumed.
• CMM probing of accessible bores: For through-bores and counterbores a probe can reach, CMM adds dimensional accuracy to accessible internal geometry. Not useful for blind passages or intersecting galleries.
• Industrial CT scanning: Full volumetric reconstruction including all internal walls. A Nikon XT H 225 CT system - rated to 225 kV, standard in precision industrial CT labs - delivers voxel resolution ranging from approximately 0.003-0.104 mm depending on specimen diameter, with steel penetration capability of approximately 30 mm. For smaller aluminum castings under 150 mm where thin-wall passages (< 3 mm) need to be resolved reliably, higher-energy benchtop systems such as the GE Phoenix v|tome|x m with the 240 kV tube are appropriate. For larger steel housings requiring deeper penetration, higher-kV industrial CT systems (e.g., 450 kV class) are available at specialist labs and provide coarser voxel resolution in exchange for the penetration capability - confirm part size and wall thickness with the CT lab before committing. CT produces a complete point cloud of every surface, external and internal, in a single acquisition.

Representative CT cost ranges by part type (confirm current pricing with your chosen CT lab):

For a complex hydraulic valve body with multiple intersecting oil galleries - say, a 200 mm gray iron manifold block with five intersecting 12 mm passages - CT resolves the passage geometry, core shift, and wall thickness at every intersection. A pattern built from incomplete internal geometry on a pressure-retaining part is a failure risk that no client can afford. The CT cost is a fraction of the cost of a pattern rebuild after discovering an under-thickness wall on first article.

For deep coverage of this approach, see our post on CT scanning to recover internal casting features.

Rule of thumb: any internal pressure-retaining passage warrants CT before committing to a pattern.

---

Accuracy, Tolerances & What to Specify When You Request a Quote

One of the most common client confusions is equating scan accuracy with final casting accuracy. They are not the same, and each step in the chain adds to the total stack-up.

Typical accuracy chain for casting reverse engineering:

The casting process itself introduces more dimensional variation than all upstream steps combined. For foundry pattern work, ±0.1 mm on the CAD model relative to the physical part is the correct accuracy target - tighter than that provides no benefit because the casting process cannot hold it.

For direct CNC reproduction from billet, the full stack-up matters and tighter scan accuracy is warranted: arm scan ±0.025 mm, mesh fit ±0.03 mm, CAD ±0.05 mm.

What to tell us when you request a quote:

1. Material and casting process (gray iron / green sand, die-cast aluminum, etc.)
2. Which surfaces are machined after casting - these receive extra accuracy budget and drive the GD&T callouts
3. Maximum acceptable deviation on sealing faces, flange mating surfaces, and bore IDs
4. Intended end use: new foundry pattern, mold cavity, or direct CNC replacement - each requires different shrink treatment and different CAD outputs

Deliverable tiers and pricing:

For a complete pre-submission checklist, see what to include in a reverse engineering quote submission. For the full pipeline from scanning to CAD environment, see our post on the full scan-to-CAD workflow.

We also cover the broader category of orphaned and discontinued components in reverse engineering discontinued and obsolete spare parts, and what to send when requesting a reverse engineering quote walks through exactly what we need from you.

---

FAQ

Can you reverse engineer a cast part if you only have one physical sample?

Yes - one sample is sufficient for geometry capture. We flag in the project documentation that a single part may carry wear, minor deformation, or casting defects not representative of the nominal design. Best practice is to provide both a worn and an unworn example when available, but a single serviceable part yields a usable pattern geometry. During processing we generate a surface deviation map of the scan itself; any localized anomaly - a ding, a repaired crack, a worn boss - shows up in that map. We review it with the client before committing to the CAD model, and the engineer makes a judgment call on whether to model the anomaly as-found or infer the nominal geometry from adjacent context.

How do you know what shrink factor to apply if you don’t have the original material spec?

XRF (X-ray fluorescence) spot analysis identifies the alloy reliably and non-destructively. For iron castings in the field, a spark test gives a rough carbon content indicator - white sparks indicate high carbon, consistent with gray or ductile iron. For a confirmed alloy, the foundry producing the replacement parts should always validate the shrink factor against their specific process and pouring temperature. The stakes are real: a ±0.2% error in shrink allowance adds 0.6 mm to a 300 mm part. On a bore ID feeding a shaft fit, that is the difference between a sliding fit and an interference fit.

What file format does a foundry or pattern shop need?

STEP AP214 is the universal currency - every commercial CAM system reads it without translation loss. Some shops prefer Parasolid (.x_t), equally valid. Do not deliver STL for pattern work: it carries no units in the header, has no feature tree, and toolpaths generated from a faceted mesh produce a visibly rougher pattern surface than paths generated from an analytical solid. Always pair the STEP with a 2D PDF drawing carrying GD&T callouts for all machined surfaces - the STEP alone is not sufficient for inspection or machine setup reference.

How long does it take to reverse engineer a cast valve body for a new foundry pattern?

For a part under 300 mm with no internal cored passages, 5-12 business days. Here is where that time actually goes: scan and setup 0.5 day, point cloud processing and registration 1 day, mesh reconstruction 0.5 day, hybrid CAD modeling 2-4 days (scales with part complexity), 2D drawing and GD&T callouts 1 day, internal QC and deviation report 0.5 day, client review and revision cycle 1-2 days. Complex multi-core housings or parts requiring industrial CT add the CT acquisition and reconstruction time (typically 2-5 additional days depending on lab scheduling) and extend total delivery to 2-4 weeks. If you have a hard foundry deadline, share it at quote time - we build the project schedule around it.

Can the scanned model go directly to 3D printing instead of a foundry pattern?

Yes - direct sand mold printing via ExOne or voxeljet systems prints the mold cavity directly from the STEP file, skipping the physical pattern entirely. This approach is increasingly common for short-run castings in the 1-50 piece range. Shrink must still be applied to the model regardless - the metal still shrinks. For production runs over 200 pieces, a CNC-machined aluminum pattern wins on per-part cost because the printed mold cost repeats on every pour.

What is the difference between scanning a part for a foundry pattern vs. scanning for direct CNC reproduction?

Fundamentally different outputs from the same scan. For a foundry pattern: shrink is added to the solid, draft angles are preserved as-found, and the parting line is explicitly modeled. For direct CNC reproduction from billet: no shrink is applied, draft angles may be removed because a machined billet part does not need mold pull, and the model represents the finished part geometry. These are two different CAD models from the same scan data. Delivering the wrong model to the wrong shop creates expensive rework in both directions - specify intended end use before modeling begins.

---

Put Your Cast Part Back Into Production

If you have a cast part with no original pattern or drawings, email us photos and the part’s approximate envelope dimensions. We return a scoped quote with recommended scan method, timeline, and fixed price - same business day for parts under 500 mm.

We handle gray iron, ductile iron, aluminum, steel, and bronze castings across green sand, no-bake, die cast, and investment cast processes. We deliver a complete foundry-ready package: STEP solid with shrink applied and draft annotated, 2D drawing with GD&T on all machined surfaces, color deviation report zone-by-zone, fillet radius table, and a shrink confirmation memo the foundry can sign off on before the first pour.

Request a reverse engineering quote or review what to send when requesting a reverse engineering quote to prepare your submission.