In manufacturing, an unexpected equipment failure usually costs more than the part itself: a line stops and orders are delayed. We are currently delivering an EU-funded project in Latvia in which artificial intelligence monitors industrial equipment and warns about deviations. Below we briefly describe how it is built. The project is still ongoing, so we make no promises about results, but we share the architecture and the lessons.
The task
The equipment already has controllers and sensors, but the data is invisible or nobody watches it continuously. The system must notice by itself when the equipment’s behaviour starts to deviate from normal and tell a person clearly what to do.
How the solution works
- Data: signals from equipment controllers and sensors are read continuously and stored in a time-series database.
- Deviation detection: a statistical model learns what normal operation looks like and notices deviations or trends that may indicate wear.
- A clear message: a local AI model writes a human-readable alert text and another translates it into the local language.
- Email alerts: operators rarely watch a dashboard, so the alert arrives where they already work.
- Dashboard and access: a real-time view and invited users only, with no open registration.
Why local models
Production data is often sensitive and connectivity is not always reliable. So alert generation and translation run on local models, with no need to send data to external AI services.
What we learned
- Start with a few signals, not all of them. First where a failure is most expensive.
- Baselines must be calibrated on real equipment operation, otherwise there will be too many false alarms.
- An alert must be actionable: what changed, how far from normal and what to check. Otherwise it becomes noise.
- Language matters: an alert in the operator’s language is noticed faster.
- Simulation helps testing before real equipment is connected.
Who it suits
Manufacturers with equipment that has controllers and who want fewer unexpected stops and planned maintenance. Such projects may also fit EU funding conditions: we wrote about it in “EU funding for AI in manufacturing in Lithuania”.
If you want to discuss how to adapt such a solution to your equipment, book a free consultation. More about our AI solutions.
Want to discuss your case?
The first consultation is free. We’ll map your processes and suggest concrete next steps.