Case: 3D Scanning Helps Large Steam Turbine Castings Detect Margins and Mark Lines

A steam turbine casing is a sand casting, usually 1 to 6 tonnes and 1,500 to 4,000 mm across. Draft angles, core shift, and uneven cooling mean the as-cast surface never sits exactly where the model says. The split faces, the bore, and the flange bolt circle are the areas that carry the tightest requirements, often ±0.1 mm or better after final boring.

The machine shop gets a part with no reliable reference. Clamp it on the raw surface and the first face may cut clean while the opposite flange runs 4 mm short of stock. Move the setup to chase that flange and the bore loses its wall thickness. Scrap risk on a part this size is not a rework cost, it is a remelt and a lost month.

Traditional layout uses height gauges, scribes, and a surface plate. That works on a 200 mm bracket. On a 3,000 mm casting the plate does not exist, the operator cannot reach the middle of a curved wall, and every measurement is relative to an edge that is itself out of tolerance.

This is where the case for 3D scanning helps most. Before any metal is cut, the shop needs a map of the actual stock: how much material sits on every machined face, where the walls are thin, and which datum set makes the whole part clean up.

We use a handheld or tripod-mounted structured-light scanner with photogrammetry targets for the large volume. The part is placed on a stable stand, not a granite plate. Reference targets are stuck to the casting and to the floor around it, then a photogrammetry pass locks the global coordinate frame to within about 0.1 mm over a 4 m diagonal.

The scan itself captures the whole exterior: split faces, flanges, nozzle openings, ribs, and the rough as-cast skin. Interior bores and deep cavities are captured separately with a smaller scanner head or a bore gauge, then merged. Total capture time on a 3 m casing runs 2 to 4 hours, depending on surface finish. Painted or heavily rusted surfaces need a matte spray first.

Registration is the step that decides everything. The point cloud is aligned to the CAD model using the datum scheme the drawing calls for, not a best-fit over the whole part. A global best-fit will average every error and hide the one face that matters. We run several alignment trials and compare the resulting stock maps before choosing.

The output is a color map in 0.5 mm bands. Blue means no stock, green means roughly 1 to 2 mm, yellow 2 to 4 mm, red above that. On a turbine casing the useful question is not the average deviation. It is the minimum stock on each machined face, and whether the minimum is positive everywhere.

• 1
  Targets before scanningPhotogrammetry frames the whole part; without it, long scans drift.
• 2
  Matte spray for shiny castingsBare or machined surfaces reflect; uncoated scans show holes.
• 3
  Separate bore captureDeep ID features need a second pass and a merge check.
• 4
  Datum-led alignmentBest-fit over the whole part hides the critical face.

A color map on a screen does not help the operator standing at a 3 m boring mill. The scan has to come back to the part as physical marks. We project the machining lines, the split-face centerline, and the bolt circle onto the casting using a laser projector aligned to the same coordinate frame as the scan.

The projector throws the CAD curves directly onto the rough surface at 1:1. The operator scribes or punches along the projected line, then the projector is removed and the setup proceeds normally. Position accuracy on the surface is roughly ±0.3 mm over the projected area, which is enough to place a roughing allowance but not enough to substitute for a bored datum.

For smaller castings that will not fit under a projector, or when the shop wants a permanent record, we mark with a CNC layout machine or a portable arm. The arm probes the surface at the scan-derived coordinates and scribes the same lines. Slower, but it works on a part that is already clamped in the machine.

The marks are not the final dimension. They tell the operator where the nominal feature sits relative to the real casting. After the first face is cut, the machined surface becomes the datum and all subsequent work references it, not the scan.

Scanning measures the outside of the part. Internal porosity, shrinkage voids, and core-related defects sit inside the wall and will not show up on a surface deviation map. A wall that reads 12 mm on the scan may still have a gas pocket at 6 mm depth. For critical pressure boundaries, ultrasonic testing or radiography is the right tool, not scanning.

The scan also has no opinion on material condition. Stress relief state, hardness, and residual stress after welding are outside its scope. If the part will be machined, welded, and machined again, the scan only describes the geometry at the moment of capture.

Surface finish limits accuracy. A rough sand-cast skin scatters the light and adds noise, typically 0.1 to 0.3 mm on as-cast surfaces. Machined or blasted surfaces scan tighter, often under 0.1 mm. When the decision hinges on a 0.2 mm margin, we scan the machined faces rather than the raw skin.

Finally, a scan is a snapshot of one part. It does not tell you whether the next casting from the same pattern will behave the same way. Pattern wear, sand condition, and pour temperature all move the result. That is why we recommend scanning the first article, then sampling at a defined interval rather than scanning every piece.