Three Dimensional Inspection of Complex Cavity Parts: How It Actually Works

• 1
  What a three dimensional inspection complex cavity parts report contains
• 2
  Why probe reach decides what can be verified
• 3
  Datum strategy changes the numbers you get
• 4
  Wall thickness, deflection and the measurement trap
• 5
  Material and finish effects on cavity measurement
• 6
  Where our cavity inspection process sits
• 7
  When three dimensional inspection is the wrong tool
• 8
  Cavity geometry versus what a CMM can still check

A three dimensional inspection complex cavity parts report is not a picture of the part. It is a set of measured coordinates tied to a datum frame, plus the deviations between those points and the CAD model. On a deep pocket or an internal rib, that difference matters more than the outside profile, because the outside is usually the easy half.

The report normally carries a datum callout, a feature list, the measured value, the nominal value and the deviation. For a cavity, the feature list is where projects go wrong. If the drawing names six dimensions and the cavity has forty surfaces that control fit and flow, the report can be complete and still useless.

Three dimensional here means the inspection resolves X, Y and Z rather than a single section. A touch probe on a CMM does this point by point. A laser line scanner does it as a point cloud. A structured light system does it as a mesh. All three produce the same kind of output: coordinates in space, compared against nominal.

The engineering question is never whether the machine can output coordinates. It is whether those coordinates land on the surfaces that actually matter, at a density high enough to catch the error mode you care about. A 12-point check on a 40 mm deep pocket will miss a taper that runs from 0.02 mm to 0.15 mm across the depth.

• 1
  Datum frameThe reference planes and axes all deviations are measured against.
• 2
  Feature listWhich surfaces are measured, and to what nominal.
• 3
  Point densityHow many points per surface, which sets the smallest error you can detect.
• 4
  Deviation bandMeasured minus nominal, usually reported per axis and as a profile value.

A touch probe is a stiff stem with a ruby ball at the end. That ball has to physically arrive at the surface, at an angle that lets the stem clear the surrounding walls. This is the single hardest limit on cavity inspection, and it is mechanical, not software.

A typical straight stylus is 20 mm to 50 mm long with a 1 mm to 3 mm ball. Reach into a pocket is roughly the usable stem length minus the clearance the stem needs to avoid rubbing the wall. On a pocket 40 mm deep and 12 mm wide, a straight 50 mm stylus often cannot reach the floor without shanking out.

The fix is a star stylus, an extension, or a different measuring direction. Each one adds joints, and every joint adds a small amount of bending under contact force. Probe bending of a few micrometres is invisible on a 0.1 mm tolerance and fatal on a ±0.005 mm tolerance.

So the reachable depth is not a fixed number. It depends on pocket width, corner radius, stylus diameter and the measuring force setting. A cavity the size of a coin and a cavity 300 mm across both need a reach plan, but the plans look nothing alike.

The same physical cavity will pass or fail depending on which datum you pick. Align to the outside faces and the internal walls absorb all the machining error. Align to the internal walls and the outside faces absorb it. Neither is wrong. The choice has to match how the part assembles.

For a housing that bolts to a flat plate, the mounting face is the datum and the cavity walls are measured relative to it. For a manifold that seats on an O-ring, the sealing groove is the functional datum and the bolt pattern is measured relative to the groove. Getting this backwards produces a report that is technically accurate and commercially misleading.

We ask for the assembly drawing, not just the part drawing, before quoting a cavity part. The assembly tells us which surface touches what. That single piece of information removes most of the argument that follows a first-article inspection.

When the drawing gives no datum at all, the default is the three primary planes of the block. That is a reasonable fallback for a simple part and a poor one for a cavity with a dominant sealing face. It also means two suppliers can report different results on identical parts and both be telling the truth.

Thin cavity walls move. A wall 1 mm thick between two pockets will deflect under probe contact force, and the probe records the deflected position, not the free-state position. The part then measures out of tolerance and the machinist chases a problem that does not exist in service.

The usual fix is to lower the measuring force and slow the approach. That helps, but it also raises cycle time and may still not be enough on a 0.5 mm wall. The better answer is to fixturing the part in a way that supports the wall, or to measure it on a non-contact system.

Non-contact options are a laser line scanner or structured light. Both capture a dense point cloud without touching the surface. The trade-off is that they see optical surfaces, not true surfaces. A shiny aluminium wall, a transparent film or a dark anodised finish will each bias the result in a different direction.

The practical rule: use touch for dimensions that control fit, use non-contact for form and for walls thin enough to move. When both are used, report them separately. Mixing CMM points and scanner points into one profile number hides the fact that they carry different uncertainty.

Aluminium 6061 and 7075 cut cleanly and hold a cavity wall well. Titanium TC4 (Ti-6Al-4V) and Inconel move more during machining, so a cavity measured cold may not match the same cavity measured after stress relief. On tight cavity work we measure after the finishing operation, not before.

Stainless 316L and 17-4PH hold form but work-harden at the cut. A cavity floor that was spring-passed at the end of the cycle can carry a residual stress layer that shows up as a small bow when the part is released from the vise. Measuring in the fixture hides it. Measuring free hides something else.

Surface finish interacts directly with optical inspection. Bead blasting and tumbling scatter a laser line and add noise to the point cloud. Hardcoat anodising builds a layer that can be 20 μm to 50 μm thick, which shifts every external dimension by roughly that amount. If the drawing calls for a finished dimension, the inspection has to happen after anodising.

Plastics add a different problem. POM and PEEK are dimensionally stable but absorb moisture. PA and ABS move with humidity and temperature. A cavity measured at 20 °C in the metrology room and used at 40 °C in service will not match the report, and that is a design issue, not a machining issue.

We machine cavity parts on 16 simultaneous 5-axis centers, 12 four-axis mills and 27 three-axis machines, with a maximum processing size of 4,000 mm. A Ø400 mm rotary table handles the round cavity work that would otherwise need two setups, and two setups means two datum shifts and two chances to lose alignment.

Tolerances hold at ±0.005 mm (±0.0002 in) on features the probe can reach. Surface finish runs from Ra 0.2–0.8 μm on a fine finish to Ra 1.6–3.2 μm as machined, and the finish you need is usually set by the sealing or flow requirement, not by preference.

Every part is inspected before shipment, with raw material check, in-process monitoring and final inspection. Reports go out on request. We hold ISO 9001:2015, IATF 16949:2016, ISO 13485:2016 and ISO 27001:2022, which matters when the cavity part is a medical housing or an automotive fluid path.

The number that matters most on cavity work is the one nobody prints: how many features are actually measured versus how many the drawing calls out. We list that gap in the first article report so the customer can decide whether the remaining surfaces need a different method.

Some geometry cannot be verified by any external probe. A closed internal cooling channel, a lattice inside a solid block, or an undercut face behind a wall has no line of sight and no probe path. Adding more CMM time will not change that.

The alternatives are limited. You can section a sacrificial part and measure the cut face, which destroys it but gives a true section. You can use industrial CT, which resolves internal geometry without cutting but costs more per part and has its own resolution limit. Or you can accept process control instead of part verification: prove the first part by section, then trust the machine and the program for the run.

Process control is legitimate when the process is stable. It is a bad trade on a five-piece prototype run in a new material, where the whole point is to find out whether the design works.

The honest position is that a three dimensional inspection complex cavity parts plan should state what is not measured. A report with a clear exclusion list is more useful than one that implies full coverage and quietly samples twelve points on a 40 mm deep pocket.