The industrial world is undergoing rapid transformation as edge AI solutions become more commonplace, helping to increase operational efficiency, reduce downtime, and ensure worker safety. In a recent webinar, Edge Impulse was joined by Renesas, Syntiant, and Tria (formerly Avnet Embedded) to discuss how small microcontroller-based devices such as the Tria RaSynBoard enable on-device AI in small, low-power packages.
The Rising Importance of Edge AI in Industrial Applications
Many traditional industrial processes rely on centralized systems for monitoring and alerting on machine functionality. While an alert that a machine has stopped is certainly useful to an operator, a superior system would be monitoring other parameters that can be leading indicators of failure such as strange noises, unexpected vibration, increased temperatures, etc. If anomalies occur, downtime for investigation or repair can be scheduled in advance, at a known time, avoiding unplanned outages. Predictive maintenance practices can help reduce equipment support costs, minimize downtime, and can help ensure worker safety as well. Edge AI can also assist with on-device machine learning algorithms that can make sense of sensor data, audio, or visual inputs to add context to the operational status of equipment.
Key Topics Covered in the Webinar
Syntiant opened the conversation by discussing their Neural Decision Processors (NDPs), which are AI accelerators specifically designed for always-on audio applications, such as keyword spotting or audio classification. Syntiant’s NDP120 for example, offers 100x the efficiency and 30x the throughput of traditional MCUs when running neural networks for anomaly detection, event monitoring, and predictive maintenance.
Following Syntiant’s presentation, Tria then joined the conversation and gave a deep dive on their Renesas RA6M4-powered RaSynBoard. In addition to the RA6 MCU, there is a Syntiant NDP120 accelerator on the device for running AI models. There is also an IMU for motion interpretation, a microphone, Wi-Fi and BLE for connectivity, and an SD card slot for persistent data storage.
The final segment of the webinar was a demonstration of how Edge Impulse can be used to build a custom machine learning model, followed by a walk-through of how to deploy that model onto the RaSynBoard. A sample use-case was also shown, with a live demo of a compressor pumping water in both a normal operational state, as well as an anomaly state with reduced flow that was correctly identified by the RaSynBoard, with a visual indication of red LED lighting used to notifying a user.
Watch Now, On-demand
Although the webinar was originally presented as a livestream, you can now watch it on-demand at your convenience by simply registering here.
There are many insights presented, and a large body of information is delivered by the presenters, so if you need any clarification or have any questions, be sure to reach out to us and ask!