Broken tool detection lighting: find a drill tip in aluminum

A multi-spindle drilling line in aluminum alloy snaps a drill or a tap, and the tip stays in the part. Torque and vibration monitoring did not fit the machine, so the check has to move to the workpiece: find the steel fragment after the cut, before the part moves on or into assembly. That is mostly a lighting decision, not a sensor decision. A steel tip a fraction of a millimetre wide resolves easily on a standard area scan camera; the hard part is contrast against bright, reflective aluminum. This guide gives the decision — low-angle dark field from a ring light — the geometry and resolution math behind it, a worked example for a typical hole pattern, and the pitfalls that decide whether the check survives production.

What the camera can and cannot see

After a break the fragment presents one of three ways, and only two of them are a camera job.

1. Proud fragment or burr ring. The tip shears and part of it, or the burr it leaves, stands a few tenths of a millimetre to a few millimetres above the machined face. This is the common case, and it is what dark field is built for.
2. Tip down the bore. The fragment sits inside the open hole below flush. The camera sees it only if light reaches it, which turns the task into bore lighting.
3. Fully buried. The tip is enclosed in aluminum with nothing breaking the surface. No lighting arrangement finds it; a buried fragment needs a volumetric method such as X-ray or ultrasonic testing. Steel is ferromagnetic and aluminum is not, so a magnetic or eddy-current check is the alternative route here.

Size the target before anything else. One documented broken-drill detector design works with features from 0.025 mm to 0.76 mm; a snapped small drill or tap typically leaves 0.5-3 mm of proud material or debris at the hole mouth. Pick the smallest fragment you must catch — 0.5 mm is a defensible target for a guaranteed catch — and carry that number through the whole design.

The decision: low-angle dark field

Light the surface at a grazing angle and let geometry make the contrast. A dark-field ring or bar aims its rays 15-30° above the machined face. The flat aluminum reflects that light away from the camera and reads near black. Anything standing proud — a steel tip, a burr, a chipped edge — catches the rays on its walls and scatters them back bright. The inspection collapses to a bright blob on a dark field: a threshold and a blob count instead of a fragile gray-level comparison.

Tone contrast alone is the wrong tool here. Polished aluminum reflects roughly 85-90% of visible light and steel roughly 55-60%, so both read bright against a dark background, and the tone gap is destroyed by cutting oil, machining marks, anodizing and exposure drift. Geometry contrast survives all of that: the flat face stays dark whatever its finish, because the specular reflection simply leaves the frame.

Two consequences for the set-up. Take the lowest angle your light offers and walk it up: lower angles exaggerate relief, while below about 10° the machining marks themselves scatter and the field stops reading black. And because a raised steel fragment is usually the brightest thing in the frame, a mono camera with a fixed threshold is enough.

For the cases where dark field loses to a dome or a backlight, see when to pick a dome, dark field or backlight. A zoned ring read from several directions is also the basis of photometric stereo ring lighting for surface inspection, which makes surface relief even more explicit.

Lighting geometry: angle, direction and coverage

The first line sets the angle. A fragment standing h proud, lit at elevation α, shadows a strip s of flat surface behind it, and its lit wall is what the camera sees. At h = 1 mm: 15° gives s = 1 / tan(15°) = 3.7 mm, 30° gives 1.7 mm, 45° gives 1.0 mm. The deeper the grazing angle, the longer the signature — which is why the low end of the range wins until texture noise takes over.

Direction is the second variable. You cannot know which way the fragment walls face, so light from all around: a ring, or a zoned ring fired quadrant by quadrant. With a multi-channel sequencing source such as the LS-SPC-0086 you fire one zone per pass and flag the part when any pass shows a blob at a hole edge.

Coverage and timing: mount the ring so its inner rim sits just outside the field of view, leaving 20-40 mm of radial clearance at a working distance near 100 mm so the housing never shades the grazing rays. Fire the light as a strobe of 0.1-1 ms from the machine cycle-complete signal: the strobe kills ambient light, freezes residual motion and keeps duty cycle low. Overdrive limits and duty-cycle math live in strobe overdrive, duty cycle and timing; if the first frames after a trigger come back black, the cause is usually light timing rather than exposure — see why the first camera frame comes out dark.

Worked example: a 0.5 mm tip beside a 6 mm hole, 40 × 30 mm window

Inputs, each with its source:

The rules: pixel scale is field of view divided by pixels, and the fragment must span at least 4 pixels; magnification is sensor width over field of view; focal length follows from f = WD × m / (1 + m).

The arithmetic:

• Pixel scale: 40 mm / 4000 px = 0.010 mm/px. The fragment spans 0.5 / 0.010 = 50 pixels, more than 12 times the minimum.
• Magnification: m = 7.4 / 40 = 0.185.
• Focal length at 100 mm working distance: f = 100 × 0.185 / 1.185 = 15.6 mm, so the nearest standard lens is 16 mm.
• Check at 16 mm and 105 mm: m = 16 / (105 - 16) = 0.180, FOV = 7.4 / 0.180 = 41 mm. The window fits with about 0.6 mm of margin per side, the scale becomes 41 / 4000 = 0.0103 mm/px, and the fragment still spans 0.5 / 0.0103 ≈ 49 pixels.
• Signature check at 1 mm proud and 20° elevation: the shadow strip is 1 / tan(20°) = 2.7 mm long, about 270 pixels at this scale.

Decision: resolution is not the limiting factor. The camera puts roughly 50 pixels on the 0.5 mm target, and on this sensor the 4-pixel budget would hold for windows up to 500 mm wide (4000 px × 0.125 mm/px). Spend the effort where the risk is: a low-angle ring at 15-30°, strobed, with the flat field proven dark on a known-good part. If the window changes, recompute f with the focal length calculator below.

Camera, lens, light and trigger set-up

A mono camera is enough — you are thresholding brightness, not judging colour; for the wavelength question anyway, see which LED colour gives the best contrast. Shutter type follows motion: rolling shutter is fine because the part is clamped during the grab; choose a global-shutter model only for imaging on the fly. Exposure is arithmetic, not guesswork: keep blur under a pixel, and blur = speed × exposure, so the 100 mm/s moving part allows 0.0103 mm / 100 mm/s = 0.10 ms of exposure — strobe territory — while a clamped part tolerates tens of milliseconds.

The lens follows from the worked example: 16 mm at about 105 mm working distance for that 40 × 30 mm window, recomputed for any other window with the calculator above. If the tip sits below flush inside the hole, a grazing ring cannot reach it: at 20° elevation the rays cross a 6 mm bore and drop 6 × tan(20°) = 2.2 mm, lighting little more than the top 2 mm of the far wall. Hand that case to a bore layout — cylindrical bore inspection lighting and camera layout — and for shallow recesses in general, lighting for bright, recessed contacts.

Trigger wiring: the light controller and the camera both take the machine cycle-complete signal; on a zoned ring, grab one strobed frame per zone; pass and fail come from a blob check gated to the hole-edge regions of interest.

Pitfalls that decide whether it works

In the order they usually bite:

1. Swarf glitters like the fragment. Fresh aluminum chips are raised, angled scatterers — exactly what dark field lights up. Blow the window clear with an air jet, and gate every blob to a hole-edge region so a chip elsewhere cannot fail a good part. The cleanup tools are the same as in dark speck detection with thresholding and morphology.
2. Stray light lifts the dark floor. Overhead lighting reflects off the aluminum and shrinks the contrast margin. Enclose the station or lean on the strobe, and chase persistent wash-out with the stray-light low-contrast checklist.
3. Flush breaks leave nothing proud. When the tip shears flush, the signature is the burr ring or crater at the hole edge, not the fragment — inspect an annulus around each hole rather than the hole itself.
4. Anodized or coated parts. Tone contrast dies, geometry contrast does not; on black anodized aluminum the steel fragment remains the bright exception, so keep the same lighting.
5. Fragments deeper than the light reaches. More than a couple of millimetres down the bore, or fully buried in the face, optics is out: bore lighting for the first case, X-ray, ultrasonic or a magnetic check for the second.

Where the camera station sits among detection methods

Post-process inspection layers on top of what the machine can know; it does not replace it. Controls usually raise an alarm or a coded event on a break, and a tool eye or touch probe can measure the tool after the cycle — this drill tool length offsets explainer on tip versus shoulder measurement shows why the measurement point changes what you learn. The in-machine options are surveyed in broken tool detection methods and codes for CNC lathes. The camera answers the question those methods leave open: did a fragment end up in this particular part? Checking after each operation also catches the failure at its cheapest moment — before the next tool machines into a part that already carries steel.

Parts list and set-up checks

Camera. The LS-CMR-0166 is a 12 MP USB3 camera with an IMX226 1/1.7" rolling-shutter sensor, 4000 × 3000 resolution at 30 fps.

Lights. The LS-SPC-0142 is an Ø80 mm multi-angle ring on 24 V, the LS-SPC-0086 a 150 mm 8-zone sequencing source on 24 V, the LS-SPC-0129 a 48 V high-power strobe for faster or longer-stand stations.

Match the ring to the window: the 80 mm ring suits single-hole checks, the 150 mm zoned source covers the 40 × 30 mm example with a four-direction pass, and the strobe takes over when the part moves during the grab.

Before you switch the station on:

1. Confirm the smallest fragment you must catch, in mm, and recompute the pixels it will span.
2. Shoot a known-good part under the final lighting and count the blobs; that count is your false-positive floor.
3. Seed a real defect: glue a cut steel pin of the target size beside a hole and confirm the station fails it.
4. Sweep every angle the ring offers and keep the one with the highest fragment-to-background contrast on the seeded part.
5. Re-check the dark floor with the room lights on and off; enclose the station if it moves.
6. Gate blobs to hole-edge regions so swarf elsewhere never stops a good part.

To go further, pick the camera on the camera finder, match the light on the light finder, or start the configurator at the light step; the area scan cameras and special lights categories hold the parts, and the lighting topic page collects the neighbouring set-ups. If the part is awkward — deep bores, curved castings, coolant everywhere — describe it on the contact page and we will help you spec the station.