KOMPASS AI brings scratch detection to your production line. It inspects every part under lighting set up for your finish, flags hairline and low-contrast scratches in real time, and triggers the reject before the part moves on to packing, plating or assembly.
Scratch detection depends on light. Tilt a machined part under a lamp and a shallow scratch appears. Tilt it back and it disappears into the finish. Inspectors find scratches this way, one part at a time. At line speed, nobody can tilt every part.
Our AI Product Inspection system, KOMPASS™, paired with modular vision hardware, checks every part for scratches, learns what acceptable finish looks like on your parts and keeps an image of every part it passes or rejects.
A polished rod can look flawless while reflections hide scratches, dents, vibration marks and grinding marks. Watch KOMPASS inspect the full circumference of each rod, classify surface defects in real time and separate the rejects automatically.
Cameras and lights are positioned for your part and finish, with low-angle and multi-angle lighting where a shallow scratch needs it.
The model separates scratches from polishing lines, machining marks and glare, then records each defect's type and location on the part.
A scratched part triggers the reject gate. Its image is stored with the defect location, so your team can trace the cause and fix it at the source.
Send us a scratched part and a good one. We will show you what KOMPASS sees on your finish.
Get In TouchScratch inspection moves from a judgement call under a lamp to the same check on every part, with an image behind every reject.
How Shriji inspects the top and inner surface of every cap on its wad assembly line.
Single-camera cap inspection with automatic rejection:
Scratch detection is the automated check for linear surface damage on a part, such as hairline marks, scuffs and deeper gouges left by handling, machining or assembly. Cameras capture each part under controlled light, and a trained model decides whether a mark is a scratch, where it sits and whether it breaks your acceptance rule. On a production line, it runs as one step of automated visual inspection, so every part gets the same check at line speed.
The limit is set by pixel size and lighting. As a working rule, the narrowest scratch you need to catch should span at least three to five pixels, so the field of view and sensor resolution decide what is possible. Hairline scratch detection on a large part often needs more than one scratch detection camera or view. During the line audit, Jidoka engineers size the cameras and optics to the smallest scratch in your specification.
Lighting decides whether a camera can see the scratch at all, because a shallow scratch mainly changes how the surface reflects. Under flat bright-field light it returns almost the same brightness as the finish around it. Dark field illumination sends light in at a low angle, so the intact surface looks dark while the scratch scatters light back to the lens. Machine vision lighting for glossy parts often combines several angles or patterns, confirmed on your parts before go-live.
Accuracy depends on how well the imaging reveals the scratch and how clearly the acceptance rule is written. With suitable lighting and a model trained on your parts, KOMPASS achieves over 99% detection accuracy in validated applications. Reflective surface inspection on polished metal needs the most care, since glare can hide a scratch at one angle and reveal it at another. Validate on your own good and scratched parts, including borderline cases, before the system rejects parts on the line.
It learns the difference from examples. Brushed, blasted, polished and moulded finishes vary from part to part while staying in specification, and a fixed threshold tight enough to catch a faint scratch will also flag that variation. Training on good parts that cover the full range of acceptable finish teaches the model where the boundary sits. That is how AI-based surface finish inspection brings down the false reject rate, with KOMPASS benchmarks showing 40% fewer false rejections. Keep a set of borderline parts to review any change to the acceptance rule.
Yes. KOMPASS inspects parts in-line and triggers reject gates within the line's dwell time, at up to 12,000 parts per minute depending on the part, the number of views and the imaging setup. Shriji, for example, inspects pharma bottle caps at more than 800 parts per minute with automatic rejection. Offline or benchtop scratch inspection still suits laboratory audits and first-article checks. On a running line, inline inspection catches a scratch at the station where it happens, so the cause can be fixed before more parts are affected.
Yes, with the right hardware. Rods, shafts, caps and bearings are rotated in front of the camera or captured from several angles, so the whole surface is covered. Surface scratch detection on a mirror-finish rod needs the rotation and lighting planned together, because the reflection moves as the part turns. The modular vision hardware is configured to your part geometry during deployment, and each view is validated before the line relies on it.
Yes. The same KOMPASS model classifies scratches alongside dents, chips, stains and foreign particles, so one inspection station can cover the cosmetic defects your specification lists. Each defect type is trained and validated with its own acceptance rule. Scratches are one defect class within surface defect detection, and the Defect and Surface Inspection page covers the full range. Tell us which defects your customers reject most often, and we will scope the inspection around them.
