Bringing AI onto the factory floor can quickly become expensive.
An engineering team might begin by prototyping an application on one platform, move to different hardware for testing, and then have to migrate everything again when it is time for deployment. Each stage adds new hardware, development work and potentially another software environment to learn.
The OnLogic Factor 101, or FR101, takes a different approach. It is designed to give engineers a compact industrial edge AI platform that can be used from the early stages of development through to the final deployed application.
We wanted to find out how accessible that process actually is, so Sandro Mark took on the challenge of deploying an AI model to the FR101 with no previous experience using the platform.
Edge AI built for industrial environments
At the centre of the FR101 is Qualcomm’s QCS6490 processor, combining eight Kryo 670 CPU cores with a dedicated NPU capable of 12 dense TOPS of AI performance. The system is configured with 8 GB of LPDDR4 memory and is designed for running AI workloads locally at the edge.
That local processing can be particularly useful in industrial environments where sending data to the cloud may introduce privacy, connectivity or latency concerns.
The FR101 also provides connectivity for integrating AI into real machinery and infrastructure. Alongside Gigabit and 10 Gigabit Ethernet, USB connectivity and USB-C, the system includes an eight-channel digital I/O interface with four inputs and four outputs.
Those interfaces could allow an AI application to do more than simply generate an inference result. For example, a vision model could detect a safety condition and use the digital outputs to interact with equipment on the factory floor.
The system is also fanless and compact, weighing around 520 grams, making it suitable for installations where a conventional workstation or server would be impractical.
From a simple model to working inference
Specifications are useful, but the important question is how easily an engineer can actually get an application running.
To test that, we trained a simple FOMO object detection model using Edge Impulse. The model was designed to identify dice within a live camera feed and was then deployed onto the FR101.
Before deployment, the device needed the Qualcomm AI Runtime SDK and Edge Impulse command-line tools installed. Once configured, the Edge Impulse Linux Runner could connect the FR101 to an Edge Impulse account, display the available projects and download the selected model directly onto the device.
From there, the model could run locally on the FR101 and provide a live video stream showing inference results.
Despite being deliberately created as a straightforward demonstration by someone without previous experience on the platform, the application was successfully deployed and running on the device.
Develop locally or connect remotely
The FR101 can also be accessed remotely using SSH.
After finding the device’s IP address, we connected to it from a laptop on the same network and were able to execute the same terminal commands remotely.
That becomes particularly useful once systems move away from the development bench and into real installations. Engineers can continue accessing and managing the device without needing a dedicated display, keyboard or other peripherals connected directly to the system.
One platform throughout development
The most interesting aspect of the Factor 101 is not any single specification. It is the possibility of keeping the same hardware platform throughout more of the development cycle.
Instead of creating a proof of concept on one system and then migrating the application onto dedicated industrial hardware later, an engineer could potentially begin experimenting on the FR101, develop and validate the application, then deploy the same platform into the final environment.
For teams trying to introduce AI into industrial systems without adding unnecessary hardware migrations, that can simplify both the technical process and the overall cost of development.
Our experiment was intentionally simple, but it demonstrated the basic proposition. We started with an AI model, configured the Factor 101, deployed the model and had inference running locally on an industrial edge computer.
Watch the full video to see Sandro Mark take the OnLogic Factor 101 from setup to a working edge AI application.
Comments are closed.
Comments
No comments yet