Glossary · AI engineering and governance
Embedded AI
Also known as: On-device AI, TinyML
German: Eingebettete KI
In industrial AI, embedded AI is the integration of AI models directly into a device's firmware or hardware, such as a smart sensor, camera, drive or controller, running on microcontrollers, digital signal processors or dedicated AI accelerators with limited memory and power.
- Industrial AI
- AI
In one sentence
Embedded AI integrates AI models directly into a device such as a sensor, camera or drive, running on limited on-board hardware.
Example
A vibration sensor with an embedded model classifies bearing condition on the device and transmits only the status and a health score over IO-Link.
How it applies
- Engineering: Embedded AI requires small, efficient models and careful verification on the target hardware. The model becomes part of the device firmware.
- Maintenance: Updating an embedded model means a firmware update. Define how model updates are distributed, verified and rolled back (Update capability).
- Documentation: The device documentation should state that an AI function is included, what it outputs, under which conditions it was validated and how its output should be interpreted.
Embedded AI vs. edge AI
Embedded AI runs inside the field device itself. Edge AI runs on a separate computer near the process. Embedded AI has stricter resource limits but can reduce data traffic to a minimum. Because the model is hidden inside the device, users often cannot inspect it, which makes clear documentation of its behavior all the more important.