AI applied to industrial PLCs
What it means
AI applied to a PLC means running machine learning models on data from the process the PLC controls: detecting a pattern that does not look normal, checking a product with a camera, or deciding when an alert is worth sending. The PLC keeps doing the control; the model adds analysis.
Three ways to add AI to a control system
| Approach | How it works | Trade-offs |
|---|---|---|
| Cloud | The PLC or a gateway sends raw data to a cloud service that runs the model | Needs a stable connection, sends production data off site, adds latency |
| Edge PC or gateway next to the PLC | A second device in the cabinet runs the model and talks to the PLC | Two devices to power, wire, update and maintain; integration repeated on every project |
| AI inside the PLC | The same device runs control on one processor and the model on another | One device; data stays on site; model size limited by the device RAM |
Design principles
- Keep control and AI on separate processors, so a failing AI application does not stop the process.
- Use model results as information or alerts, not as safety logic. Safety functions belong in a certified safety controller.
- Process data where it is generated and send only results, to reduce bandwidth and keep production data on site.
- Choose RAM for the workload: heavier models, vision or several containers need more memory.
Typical applications
- Predictive maintenance: anomaly detection on vibration, current or temperature of pumps and motors.
- Vision quality control: camera inspection next to the line.
- Energy monitoring: detect consumption that is out of the normal range.
- Remote sites: watch tanks, wells or stations and alert only when something changes.
Industrial Shields options
- Ardbox Q: AI inside the PLC. The first industrial PLC with Arduino UNO Q, with an STM32U585 microcontroller for real-time control and a Linux processor for data and AI models in one DIN-rail device. 19 or 20 I/O, 2 GB or 4 GB RAM. From €225.00 excluding VAT. Product page: /ardbox-q-arduino-uno-q-plc-ai
- Raspberry PLC: Linux PLC based on Raspberry Pi, with more I/O, two wired Ethernet ports and more RAM options, for projects that run AI and data services on Linux.
Frequently asked questions
Can a PLC run AI models?
A PLC with a Linux processor can. Ardbox Q runs AI models on its Linux side while a separate microcontroller runs the control program.
Does AI replace the PLC control program?
No. The control program keeps controlling the process. The model analyses data and produces alerts or information.
Does the data need to go to the cloud?
No. With AI inside the PLC, data is processed on the device and only results need to leave the site.
What happens if the AI application fails?
On Ardbox Q, control and AI run on separate processors, so the control program keeps running.