Vibration-based condition monitoring for conveyor systems, built from silicon to server.
In most SME manufacturing plants, bearing failures are still caught by ear, by smell, or by a machine grinding to a halt on the line. By the time someone notices, the damage — and the downtime — is already done. I wanted to build something that closes that gap: a system small and affordable enough for a plant manager without a dedicated reliability engineer to adopt, but rigorous enough to catch a developing bearing fault weeks before it becomes a failure. This started as an answer to that problem, and as a personal challenge to build a genuinely end-to-end product — not a demo, but a system I would trust to run unattended on a factory floor.
Predictive maintenance sits at an unusual intersection: it demands both embedded systems engineering and cloud infrastructure, and neither half can hide behind the other. On the hardware side, a small microcontroller reads vibration data from a sensor mounted on the machine, processes the raw signal on-device, and turns it into something meaningful about the machine's health — all while working within tight memory and power budgets that leave little room for error.
On the cloud side, that data has to travel securely and land somewhere it can be trusted. Every device proves its identity before it's allowed to send anything, and its readings are then picked up, validated, and stored in a database built for handling continuous streams of time-stamped data. The whole system runs on infrastructure I manage myself, which meant learning how to deploy, secure, and monitor a live service rather than just write code that runs once on my laptop.
The pipeline currently runs sensor to database end-to-end. The next layer is the interesting one: teaching the system what "normal" vibration looks like for a given machine, so it can recognize the earliest signs of something going wrong — long before a human would notice. Longer term, I'm planning hardware improvements and additional sensors that will let the system pinpoint not just that something is wrong, but which part of the machine is failing and why.
It's a project that has forced me to be a firmware engineer, a systems administrator, a database designer, and eventually a data scientist, often in the same afternoon. That's exactly why I wanted to build it.