Glossary · AI engineering and governance
Edge AI
Also known as: AI at the edge, Edge inference
German: Edge AI
In industrial AI, edge AI is the execution of AI models, mainly inference, on edge devices close to the machine or process rather than in the cloud, to reduce latency, bandwidth and dependence on network connections and to keep data on site.
- Industrial AI
- AI
In one sentence
Edge AI runs AI models on edge devices near the machine, reducing latency and bandwidth and keeping data on site.
Example
A vision model running on an edge device with a GPU inspects 30 parts per second on a conveyor and sends only defect images to the central server.
How it applies
- Engineering: Models often have to be compressed or quantized to run on edge hardware. Validate the optimized model, not only the original one, since accuracy can change.
- Operation: Many edge devices run the same model. Central management of model versions, updates and monitoring is needed to know which version runs where (Model versioning).
- Documentation: Document hardware requirements, model version, input data, performance limits and behavior when the device fails.
Edge AI vs. embedded AI
Edge AI runs on edge devices such as industrial PCs or gateways near the process. Embedded AI runs directly inside a device such as a sensor, camera or drive, typically on microcontrollers or dedicated chips with tight resource limits. Training usually still takes place centrally in both cases (Training vs. inference).