EN
News
Valve Plate Inspection: 79 Through-Holes And The Tolerance Band Problem
Overview
A valve plate is a flat part with a large number of holes that all have to be in the right place. The part is small, the hole count is high, and the tolerance band is narrow enough that a single shifted drilling makes the whole plate scrap. Inspecting it by hand is slow, and inspecting it by sampling is pointless: the defect is in the drilling, so every plate is either good or not.
Big Bird Industrial (formerly Shimada Big Bird Industrial) has supplied dimensional and defect inspection systems for automotive components since 2001. This article describes how a 79-hole plate is inspected in one pass, why the tolerance band rather than the hole count drives the optical design, and how to write acceptance criteria that survive an audit.
1. Why Hole Count Drives The Architecture
A plate with three holes can be inspected with a single camera and a fixture. A plate with seventy-nine holes cannot, for a simple reason: the field of view needed to see the whole plate at the resolution needed to measure a hole position is larger than any single camera can deliver at that resolution.
The practical answer is a multi-camera station that images the plate in overlapping tiles and stitches the measurement into one coordinate system. The alternative, a single camera on a moving stage, buys resolution at the cost of cycle time, and cycle time is the second constraint.
Once the architecture is multi-camera, the calibration problem becomes the main engineering task. Every camera has its own distortion, its own pose relative to the plate, and its own measurement uncertainty. The system is only as good as the stitching, and the stitching is only as good as the calibration artefact.
2. What Has To Be Measured On Each Hole
Three quantities matter per hole, and they have different tolerances.
Position
The centre of the hole relative to the datum features of the plate. This is the tolerance that decides whether the plate seals and whether the mating part bolts up.
Diameter
Measured as an equivalent circle or as a minimum inscribed circle, depending on the drawing. The choice matters on a punched or drilled hole with a slight taper: the two definitions can disagree by more than the tolerance.
Form
Roundness and burr presence. A hole that passes position and diameter but carries a burr on the exit side will not seal, and the burr is exactly what a two-dimensional top view cannot see.
3. The Tolerance Band Decides The Optics
The rule is simple and often ignored: the measurement uncertainty must be a fraction of the tolerance, not a fraction of the part. If the position tolerance is plus or minus 0.05 millimetres, a system with 0.03 millimetres of uncertainty is not capable, because it will reject good parts and accept bad ones with equal probability.
| Tolerance | Required uncertainty | Implication |
|---|---|---|
| Position ±0.10 mm | ≤0.02 mm | Standard multi-camera, telecentric lenses |
| Position ±0.05 mm | ≤0.01 mm | Telecentric optics, rigid fixture, thermal control |
| Diameter ±0.02 mm | ≤0.005 mm | Backlight, sub-pixel edge fitting, artefact calibration |
Two of these requirements are optical and one is mechanical. A plate that is not held flat and repeatable will move between the calibration and the measurement, and no amount of sub-pixel edge fitting will recover the lost accuracy.
4. Cycle Time: Where The Seconds Go
A 79-hole inspection does not take long per hole. It takes long per plate, and the time is dominated by three things: image acquisition, image processing and handling. Acquisition is bounded by exposure and by the number of tiles; processing scales with hole count and with the complexity of the fitting; handling is whatever it costs to load, locate and unload the plate.
The useful engineering move is to overlap them. Acquisition of tile two can run while tile one is being processed, and the plate can be located while it is still moving into position. A system designed as a strict sequence will be several times slower than one designed as a pipeline, at the same hardware cost.
5. Acceptance Criteria For A Multi-Hole Plate
A dimensional inspection acceptance test is not a demonstration. It is a measurement of the system against a known artefact, repeated enough times to state an uncertainty.
| Item | Method | Pass condition |
|---|---|---|
| Accuracy | Calibrated artefact, 20 repeats | Mean deviation within stated uncertainty |
| Repeatability | Same part, 30 load cycles | Range below one third of tolerance |
| Correlation | CMM on ten parts | Deviation within agreed band |
| Cycle time | Timed production run | Takt met with margin |
6. Where These Systems Fail In Practice
The failures are rarely optical. A fixture that wears will move the plate by more than the tolerance within a year. A calibration artefact that is measured once and stored will drift with temperature. A lighting unit that ages will shift the edge threshold and change the measured diameter by a few microns, which is enough on a tight tolerance.
The defence is a maintenance schedule that treats calibration, fixture condition and illumination as consumables rather than as commissioning tasks. A system that is verified monthly against an artefact will drift slowly and visibly; one that is verified once will fail silently.
7. When The Plate Is Part Of A Stack
A valve plate rarely works alone. It sits between a gasket, a body and a cover, and the holes that matter are the ones that must line up through the whole stack. That changes the inspection question from is this hole in tolerance to will this hole line up with the part above it, and those are not the same question.
The practical consequence is that the datum for a stacked assembly should be shared. If the plate is measured against three locating holes but the assembly is built from an outer edge, the two coordinate systems drift apart over a production run, and the plate can be perfectly in tolerance while the stack misaligns.
Stack thinking also changes what counts as a defect. A burr on the exit side of a hole that faces a gasket will crush the gasket and create a leak path, while the same burr facing an open cavity is harmless. Two identical plates can therefore have different acceptance depending on orientation, and the station must know which face is which.
Finally, stacking imposes a flatness requirement that a standalone plate inspection often omits. A plate that is in tolerance hole by hole but bowed by a tenth of a millimetre will not seal, and the resulting leak is attributed to the gasket rather than to the plate. Measuring flatness in the same station, with the plate in its free state, closes that gap for the cost of one more sensor.
8. Inline Cell Or Standalone Cell
The placement decision is usually made on floor space and it should be made on data. An inline station inherits the line's takt and its handling, which means no extra loading labour and no queue; a standalone cell can run at its own speed and can inspect parts from several lines, which matters when production volumes are uneven.
The deciding number is how many parts per hour need dimensional confidence. If every plate must be verified, inline is the only economic answer because the handling is already paid for. If a sample per batch is enough, a standalone cell with a better fixture and more time per part will deliver lower uncertainty for less money.
A practical hybrid is also common: an inline station that measures the critical few features on every part, and an offline cell that audits the full drawing on a sample. The inline station catches drift, the offline cell proves capability, and neither has to be a compromise.
Frequently Asked Questions
Can one camera inspect all 79 holes?
Only if the tolerance is loose enough that a wide field of view still resolves the edge to a fraction of the tolerance. In practice, tight tolerance and high hole count pull in opposite directions, which is why multi-camera stitching is standard.
Is backlighting always better for hole measurement?
For diameter, usually yes, because the silhouette gives a clean edge. For burrs and surface condition, no: a silhouette hides everything on the surface, so a second imaging mode is needed if burrs must be detected.
How often should the system be recalibrated?
It should be verified at a defined interval against an artefact, with recalibration triggered by the verification result rather than by the calendar alone.
Can the same station check assembly afterwards?
Yes, if the plate is presented in the same fixture and the camera can see the assembled features. Combining dimensional and assembly checks in one station saves a handling step, which is usually the largest single time cost.
Conclusion
A 79-hole plate is a measurement problem disguised as an inspection problem. The hole count sets the architecture, the tolerance band sets the optics, and the fixture sets whether either of them survives a year of production.
Specify the uncertainty first, then the cycle time, then the hardware. Systems built in that order are accepted on the first try; systems built the other way are argued about for months.
