Edge AI and IIoT

What is edge AI, and when do you need it

A sensor, a camera or a PLC creates data. With edge AI, a computer processes that data on the spot, before anything goes to a server or the cloud. That can be at a production line, but also at a pumping station, a charging site or an installation on a roof.

What it is

NIST describes edge computing as the layer of the devices themselves, with computing capability on or near a sensor, a meter or another device. That layer is also called the IoT network. Edge AI is a trained model running in that layer. It reads measurements or images, gives a verdict and only passes that verdict on.

Who does what

The device or the PLC measures and controls the machine.
The edge computer next to it collects the data, checks that it is right and runs the model.
The server or the cloud stores the results, compares sites and trains new versions of the model.
Your dashboard or ERP receives the alerts and the figures that matter.

When it pays off

Something has to happen within a fraction of a second, such as taking a product off the line.
There is too much data to send. You do not send a camera that films all day over a 4G connection.
The connection sometimes drops, and a remote installation has to keep working.
The data may not leave the site, because of customer agreements or security.

When you do not need it

A temperature you read once a minute and that nobody has to act on right away, you simply send to a server. An edge computer is then one more device to maintain, without any return.

What it takes

The model is the smallest part of the work. Most of it is around it.

Hardware that can handle the environment. Industrial edge computers mount on DIN rail in the cabinet and run on the 24V that is already there.
A connection to what is there, through OPC UA, Modbus, MQTT or plain 24V signals.
Remote updates, so nobody has to walk past twenty cabinets with a laptop.
Monitoring, so you notice when a device or sensor goes quiet.
Security for every device on the network.

Frequently asked questions

Is edge AI the same as IoT?

IoT is about devices that send data. Edge AI is what you do when you process that data on the device or next to it. Many IoT applications do not need it.

Which models run on an edge computer?

Usually small models for one task: spotting deviations in a series of measurements, judging a product on an image or predicting a value. A large language model needs more computing power and usually runs on a server.