There is a wide gap between making one prototype and manufacturing at scale, and a lot of people live in it: small teams and one-person operations turning out dozens or hundreds of boards a month. Inspection at that volume is done by eye, with a magnifier, late at night, which is exactly when a missing 0603 resistor gets missed.
Mukesh Sankhla has built a very affordable automated PCB inspection station for that gap, documented on Hackster and written up by Arduino this week.
The mechanism
A turntable with six stations sits under an overhead-mounted webcam. A servo rotates the platform, bringing each slot under the camera in turn; each board is analysed, and the table advances.
The entire parts list:
- An Arduino UNO Q board
- A webcam
- A servo motor
- An LED
- 3D-printed structure
That is it. There is no machine-vision camera, no telecentric lens, no ring light costing more than the boards being inspected — the components an industrial automated optical inspection machine is largely made of, and the reason those machines start in the tens of thousands.
Why the UNO Q is doing the work
The board choice is what makes this one device instead of two. The UNO Q carries both a microcontroller and an application processor, and this project uses each for what it is good at:
- The STM32 handles servo movement — real-time, deterministic, no operating system in the way of the pulse timing
- The Linux side runs the vision models
Split across two boards, that is a project with a serial protocol, a synchronisation problem and two things to power. On a dual-brain board it is one device where the motion control and the inference are in the same enclosure, and the awkward part — getting a Linux process and a hard-real-time loop to cooperate — is handled by the board’s own architecture rather than by your firmware.
Arduino’s own walkthrough of the camera-to-inference path on the UNO Q — the pipeline this project builds on, not the inspection rig itself.
The vision side
Detection uses YOLO models deployed through Edge Impulse. The system identifies specific components, checks for anything missing, and flags defects.
Component presence/absence is the right problem to point an object detector at, and it is worth being clear about why. YOLO localises and classifies objects in an image — so “is there a capacitor at this location” is close to its native question, and a board with a known layout gives you strong priors about where each answer should be found. The hard problems in professional AOI are the ones this does not claim: solder joint quality, tombstoning, insufficient paste, hairline bridges. Those need controlled lighting geometry and resolution that a webcam does not have.
Within its scope, though, missing-component detection catches a large share of the failures that actually reach a small builder’s bench, and it catches them before the board is populated into something.
Edge Impulse is the pragmatic choice for the deployment path: training a detector, quantising it and getting it running on an embedded Linux target is a multi-day exercise from first principles, and a platform that compresses it to an export step is why this project exists at a price a hobbyist will pay.
The part worth copying
The structure is fully 3D-printable and adaptable. The turntable is customisable to cradle whatever boards you are inspecting, and the station count is a parameter rather than a constant.
Arduino points out the consequence: four stations instead of six would let you inspect both sides of unpopulated boards and both sides of populated boards in one workflow. That is a genuinely better process than six identical slots, because it turns the fixture into a sequence — bare front, bare back, built front, built back — and catches fabrication faults before you spend components on a board that was already wrong.
That reframing is the transferable idea here, and it has nothing to do with vision models. A rotating fixture with N stations is a general machine for doing the same thing to several objects in a defined order. Swap the camera for a camera on a macro rail and it photographs small work for documentation. Swap it for a probe and it is a test jig. Swap it for a light and it is a turntable for photogrammetry.
Which, for a creative-hardware audience, is the reason to read a quality-control project: the inspection is the application, and the mechanism is the useful object.
Sources
- Building a super-affordable automated PCB inspection system — Arduino Blog
- Edge AI Based Multi Stage Visual Inspection System — Mukesh Sankhla on Hackster
- Arduino UNO Q
- Edge Impulse
- AI Upskilling with Arduino UNO Q: From Camera to AI — Arduino on YouTube