3D profile inspection of formed pressure vessel heads

A formed head arrives with a certificate of compliance and, at many shops, a cardboard template and a caliper. The certificate says the head was formed to print; the template checks one silhouette by eye. A 3D scan of the surface turns the receiving decision into a measurement: every point of the profile, compared with nominal, reported in millimetres with locations. This guide gives the tolerance arithmetic for formed heads, matches sensor types to head diameter, works a 2,000 mm ellipsoidal head acceptance end to end, and draws the line between what a scan proves and what stays on paper.

What the receiving inspection has to decide

Shops that buy formed heads for ASME Section VIII, Div. 1 vessels follow one of two routes. The first accepts the head maker's certificate of compliance and checks identity: diameter, thickness and heat number against the mill test report. The second measures on receipt, because the QC manual says so or because the service demands it. What the QC manual usually asks for is modest: a documented thickness reading at the thinnest point, and a shape check against print.

The shape check is where the paper route is blind. A tape across the opening gives diameter. A template gives one silhouette, judged by eye, and sees nothing away from its edge. A scan answers the question neither tool can: how far is every point of the surface from nominal, and where are the worst spots.

What each check proves

A template proves the head at the section it touches. A scan proves it everywhere the sensor saw, and the report itself goes in the job file. That difference is why shops with many head sizes, or jobs that need recorded shape data, move the check from the template wall to the scanner.

The tolerance and the geometry

Section VIII puts the formed-head shapes — ellipsoidal, torispherical, hemispherical, conical — under UG-32, and the drawing fixes the nominal profile. The acceptance rule that goes with templates is the long-standing 1-1/4 % of the inside diameter, measured from nominal. For a 2:1 ellipsoidal head, the nominal inside depth from the tangent line to the crown apex is D/4. A torispherical head takes its crown and knuckle radii from the drawing; the tolerance arithmetic is the same.

Two numbers come out of this: the band any single reading must sit in, and the accuracy the sensor needs. Judge every point against T, and keep the sensor's own uncertainty to about a tenth of T so the measurement does not eat the tolerance.

Worked example: a 2,000 mm ID 2:1 ellipsoidal head against a ±25 mm band

The case: a 2:1 ellipsoidal head, inside diameter D = 2,000 mm, formed from plate, received for a code vessel. The inputs:

• D = 2,000 mm, from the head drawing
• Nominal inside depth b = D/4, from the ellipsoidal geometry
• Tolerance T = 1-1/4 % of D, the template rule your QC manual writes on the form
• Readings on receipt, example measurements that show the arithmetic: apex depth 502 mm by depth gauge; scan deviations of +3 mm at the crown and −31 mm across a patch in the left knuckle

The apex reading of 502 mm sits inside the window, so the single-depth check passes the head. The scan's worst reading is −31 mm in the knuckle, and 31 exceeds 25, so that region is out of tolerance. Same head, opposite verdicts, because the gauge sampled the point where the head is best.

The decision: rework or reject, and record it against the heat number. The threshold for the form: no point of the surface may sit more than 25 mm from nominal in either direction, regardless of the apex reading. To judge a ±25 mm band without the sensor eating it, budget roughly a tenth of T — 2 to 3 mm of accuracy across the head — which millimetre-class structured-light and profiler sensors hold inside their calibrated volume.

Matching the sensor to the head

Heads that fit one capture are the fast case. The sensor takes a single shot, the software aligns the cloud to the nominal surface, and the deviation map comes out in one pass; the align-and-compare step is the one described in 3D point cloud template matching and deviation measurement. Heads past the capture volume need the sensor to travel instead. The trade-offs between the 3D methods are set out in 3D scanning method selection for inspection.

For a head that fits one shot, a projected structured-light camera does the whole job:

Past that, mount a laser profiler above the head and raster it line by line on a slide, gantry or robot. The nominal profile comes from the drawing, the pitch comes from the axis feed, and the deviation map assembles from the stacked profiles:

When the rig cannot move at all, a galvanometer-steered stereo camera sweeps the beam across the surface instead:

For very large heads inspected in quantity, wide-area line-scan 2.5D systems cover a strip per pass and suit a turning-roll set-up:

The MV-HLVS-16K100-L60 pairs a 600 mm emitting area with 16K resolution and a 120 kHz sampling rate, so one axis of motion plus head rotation builds the full map. The 3D sensors category lists the whole range, and the 3D sensor finder filters it by type.

Stand, motion and cabling

A bench set-up for profiler work is a stand with a vertical slide: the sensor hangs at its calibrated standoff while the slide sets the scan height. The LS-MNT-0040 professional slide-table stand carries 700 mm of height and a 400 x 450 mm base, enough for bench-scale heads and fixtures.

For heads of a metre and up, a robot or gantry rasters the sensor, and the requirement is that the standoff stays inside the calibrated volume in every pose — the subject of 3D camera working distance and positioning. If the rig is re-gapped between head sizes, recalibrate when the standoff changes; camera calibration depth range and accuracy: when to recalibrate gives the signs.

The pitch along the direction of travel is feed per profile. Moving the axis at 50 mm/s and saving a profile every 40 ms gives 50 × 0.040 = 2 mm between lines — twenty samples across a 40 mm dent, tight enough to catch it and loose enough to keep the readings independent. For continuous moving webs the same arithmetic runs in the line-scan camera line rate calculator.

A moving axis also moves its cable, so the runs through the drag chain need chain-rated camera power and I/O cable:

Parts for a first rig

A first rig for profile work on a 2,000 mm class head:

• One snapshot structured-light camera if the head fits the capture volume, or one laser profiler plus a slide if it does not
• A camera stand with a slide table, or a robot or gantry mount for larger heads
• Chain-rated power and I/O cable for any moving axis
• Software that registers the scan to the nominal profile and reports worst-case positive and negative deviation with positions

What the scan cannot prove

A scan reads one surface, so it says nothing about wall thickness. Thickness still comes from UT readings and the mill test report, and forming thins a head unevenly — most in the knuckle — which is why code work tracks the thickness reduction limits for formed dished heads separately from shape. Heat numbers and material traceability stay on paper too: a scan of a head with the wrong mill certificate is still a perfect scan of the wrong head.

Watch the surface finish as well. Blast-cleaned or scaled plate scans cleanly; a polished stainless knuckle at a steep angle can drop points, the same edge-loss effect that eats the edges of tall boxes at long standoff — why depth image edges vanish at long standoff walks through the mechanism. Where points drop out, the report must say so rather than interpolate across the gap.

Checks before you sign the report

• Take the nominal profile from the drawing, not from a similar size: comparing a 2:1 ellipsoid against torispherical radii makes every deviation number wrong.
• Write the tolerance on the report — 1-1/4 % of inside diameter unless the QC manual says tighter.
• Keep the sensor inside its calibrated volume at every pose, and recalibrate after any change of standoff.
• Support the head the way it sits in service and note the orientation; large thin heads deflect under their own weight, so the support belongs in the record.
• Report worst-case positive and negative deviation with their positions, filed against the heat number.
• Take the thickness readings anyway; the scan does not replace them.

When the template is still the right tool

A shop that receives one or two recurring head sizes from a supplier with a solid history, on ordinary service, is well served by a steel template, a caliper on the rim and a documented thickness check. The scan earns its place once the variety grows — past a couple of dozen sizes a template wall stops scaling — or once the job file needs recorded shape data, a new supplier needs qualifying, or the service makes you trust no certificate. The scan's nominal comes from CAD, so an added size costs a file, not a new cutout.

Where to start

Bring the head drawing — diameter, depth, shape, material and tolerance — and start the system configurator to build the stand, sensor and cable as one rig. For a rig quoted against your head sizes, send the drawings through the quote request, or ask the applications team through the contact page. The guides on the 3D vision topic page cover the capture and calibration steps that follow once the head is on the stand.