Light and Image at the Removal Station: How Light-Source Selection Decides Recognition Rate on Micro Screws

In a vision setup for a screw-removal station, the camera and lens are usually discussed first and the light source last, often treated as an accessory that just needs bolting on. For micro screws the order should be reversed: what decides whether you can see clearly is mostly not camera resolution but how the light lands.

The reason lies in the geometry of a screw head. On an M0.8 to M2.0 screw the head is roughly 1.4 to 4 millimetres across and the recess is a fraction of a millimetre wide. The camera is not resolving a large contour but a shallow depression a few pixels wide. Imaging quality there depends heavily on the angle at which light enters and reflects: at the right angle the recess wall and floor separate into light and dark with a crisp edge; a few degrees off and the whole head becomes an even bright patch or an even dark block, and no algorithm can recover it.

Industrial vision commonly uses four light forms, each with a clear domain.

Ring light is the most common, mounted coaxially around the lens. It is even, simple and cheap, and suits characters and contours on flat surfaces. On a screw head, frontal ring light flattens the depth information, because light arrives almost perpendicular and the recess floor reflects much like the surrounding face. It is good for presence detection — is there a screw — and poor for head type and damage.

Low-angle ring light, sometimes called dark-field ring light, grazes the surface at a steep angle. At that angle a flat surface reflects light away from the lens and appears dark, while slopes such as recess walls, scratches and burrs scatter light into the lens and appear bright. The recess is therefore lifted out of the image, and the deformed recess of a cam-out or the stub of a sheared head becomes clearly visible. This is the best value light for micro screws, especially on metal workpieces.

Coaxial light sends light down the lens axis through a half mirror and accepts only the reflection that returns along the same path. That makes it extremely effective on specular parts — polished stainless, chrome plating, mirror aluminium, all of which scatter glare everywhere under ordinary light, produce clean images with sharp boundaries under coaxial light. The cost is shallow depth of field and sensitivity to tilt: a slightly tilted screw goes bright on one side and dark on the other. Usable on flat, well-supported workpieces, risky on visibly deformed ones.

Backlight illuminates from behind and turns the part into a silhouette. It cannot measure the recess, but it measures presence, position and diameter very accurately because the contrast at the contour approaches the ideal. At a removal station backlight is rarely used alone, but it is valuable as an auxiliary channel: when frontal light fails due to glare or contamination, the presence judgement from backlight serves as a cross-check and prevents "cannot see" being misread as "no screw".

In real engineering one light source is usually not enough, and combinations are the norm. A typical micro-screw removal station uses low-angle ring light as the main source for the recess, coaxial light to suppress specular glare, and backlight for presence verification, with the three channels switched in sequence or acquired separately. This is not configuration padding: the three jobs have conflicting physical requirements and no single light position satisfies all of them.

Colour is another routinely overlooked variable. Wavelength changes reflectance and scattering across materials. On metal, shorter wavelengths scatter more and make surface texture more visible, which helps with fine scratches and deformed recess edges; longer wavelengths penetrate better and are less sensitive to surface contamination, which is more stable when the part carries oil or fingerprints. Many complaints that "the same machine becomes unstable when the batch changes" trace back not to the algorithm but to a change in surface finish — sandblasting replaced by brushing, for instance — which shifts the reflective behaviour so that the original angle and wavelength no longer fit.

Mechanical stability of the light source matters as much as its optical performance. If the emission angle drifts two or three degrees because a bracket loosened, the image drifts with it, and because the drift is gradual operators often notice only after yield has fallen a few points. A workable practice mounts the light on the same rigid bracket as the camera module and calibrates them together, avoiding a separately adjustable light, and writes brightness and trigger timing into the recipe so changeover switches them with the recipe instead of someone turning a knob.

Life and degradation also need managing. LED output falls slowly with hours of use, and different production batches age differently. If the station relies on brightness for its thresholds, degradation equals threshold drift. There are two engineering answers: choose a source with constant-current drive and brightness feedback to hold output steady, or put brightness into periodic checks with measured correction. The former is less work; the latter suits equipment already in production.

One level up sits the decoupling of imaging and algorithm parameters. The same product needs one parameter set before removal, to see the recess, and another after removal, to see the threaded hole; the two should be stored separately and called per process step rather than shared. Packaging imaging parameters, light channel, exposure and gain into an "imaging recipe" bound to the product model is the key step that turns a vision station from something tuned into something configured.

From a machine-design view these requirements end up as structural provisions and configuration items. In its combined fastening and removal projects, Shenzhen Honred Technology typically builds camera and light into one module on the head of its small screw removal equipment, keeps two degrees of freedom for Z and tilt adjustment, and provides an interface for a second light channel — because most projects discover during pilot production that one source is not enough, and without provision the retrofit means rebuilding the bracket.

A station built around an automatic vision screw removal machine has one more practical constraint: the light must not disturb torque judgement, and its heat must not affect nearby precision parts. A high-power ring light running continuously creates a hot zone near the lens, which is bad for thermally sensitive workpieces; strobe mode cuts average power substantially but must be strictly synchronised with camera exposure, otherwise banding appears. Details like these rarely surface at prototype stage and show up only after weeks of continuous volume production.

At the M0.8 end of the range the margin for error in light selection is smallest. The smaller the screw, the fewer pixels its head occupies, and any loss of recess contrast directly eats usable features; micro screws are also often blackened or plated, making reflectance less predictable than stainless. This is why the M0.8-M2.0 screw removal solution lists light configuration as a mandatory item on the same level as bits and torque rather than a default carried over from the previous product.

In short, light selection is not about buying a brighter lamp; it is an optical design for a specific workpiece. Decide which feature must be seen, choose the incidence angle that creates contrast for that feature, and only then settle brightness, colour and stability. Get that order right and recognition rate on micro screws stops depending on luck and on an experienced operator's feel. Get it wrong and even a high-resolution camera only photographs an illegible bright patch more sharply.