Edge AI is running machine learning models on or near the device that produces the data, such as a camera, gateway, or industrial PC, instead of in a remote cloud. It trades compute capacity for lower latency and less data transfer.
Edge AI applies edge computing to machine learning. A model is usually trained centrally, on servers or in the cloud, then deployed to hardware near the process for inference. Typical hardware includes smart cameras, industrial PCs with GPUs or accelerators, and gateways next to PLCs.
The main reasons to run at the edge are latency, bandwidth, and data control. A reject decision on a fast line cannot wait for a round trip to a data center. A vibration sensor sampling at high frequency produces far more data than a plant network should ship off site. Keeping data on premises also simplifies some security and privacy questions.
Limits are real. Edge devices have less memory and compute, so models are often compressed. Managing dozens or hundreds of deployed models, with versions, monitoring, and updates, is an operations problem in itself. Edge devices also add assets to the OT network that must be patched and secured.
Questions about Edge AI
Is edge AI the same as edge computing?
No. Edge AI is edge computing applied to machine learning. A model runs on hardware near the process, such as a smart camera, an industrial PC or a gateway, instead of in a remote cloud.
Are edge AI models trained on the edge device?
Usually not. A model is typically trained centrally, on servers or in the cloud, and then deployed to hardware near the process for inference.
What are the drawbacks of edge AI?
Edge devices have less memory and compute, so models are often compressed. Managing dozens or hundreds of deployed models, with versions, monitoring and updates, is an operations problem of its own. Each device is also one more asset on the OT network to patch and secure.
Edge AI in real plants
- Audi and Siemens: Audi and Siemens brought the AI-based Weld Splatter Detection check into series production in the Neckarsulm body shop; it detects possible weld splatter on vehicle underbodies.
- Schneider Electric and Cognex: Schneider Electric's Chasseneuil plant inspects contactors with Cognex In-Sight cameras and AI models built in OneVision, as a standard to replicate in other factories.
- Třinecké železárny and Siemens: Třinecké železárny piloted the Siemens Predictive Service Analyzer on an edge device to monitor the Sinamics S120 drives of the sizing mill at its Ostrava tube mill.
Related terms
Industrial IoT (IIoT)
Industrial IoT (IIoT) is the network of sensors, instruments, and machines in industrial settings that are connected to computers and software to collect and use operational data.
Machine vision
Machine vision is the use of cameras, lighting, and image-processing software to inspect, measure, identify, or guide automatically in industrial settings.
OT cybersecurity
OT cybersecurity is the protection of operational technology, the control systems and networks that run physical processes, against cyber threats, with safety, availability and reliable operation as the first priorities.
IT/OT convergence
IT/OT convergence is the integration of information technology, the systems that handle business data, with operational technology, the systems that monitor and control physical equipment, so data and practices flow between them.
Sources