Check out our Sensors Converge 2023 coverage.
Renesas Electronics wants you to use machine learning on the edge to protect things like air-conditioning systems (see figure). In the company’s demo, a running air-conditioning condenser was wrapped so that the system would begin to overheat. Renesas’ microcontroller was running a small machine-learning/artificial-intelligence (ML/AI) model that does a “Reality check HVAC.” Thus, it’s more than just a temperature tripwire—it’s also a self-diagnosing unit.
In the demo, Renesas employees wrapped an air-conditioning system to simulate an overheating condition that was recognized by the machine-learning application running on a Renesas microcontroller.
I talked with Stuart Feffer, Head of Real-Time-Analytics and Reality AI at Renesas, about the company’s edge AI software that targets microcontrollers.
The software is a portion of the Reality AI and RealityCheck AD for Industrial Anomaly Detection software support. Much of this was part of the Reality Analytics acquisition that brings ML solutions to edge computing, especially for small microcontrollers that are used for sensor applications and motor control.
The software support automatically generates a baseline anomaly-detection model. The model is then used to build up and train models that can predict and detect specific faults like an overheated air-conditioning condenser.
Unlike video-processing ML models that require hefty processors and often AI accelerators, this software is designed for low-power, lower-performance systems that check sensors and control motors. The small footprint models are built using ML optimization techniques like quantization, compression, and pruning. The models use small integers to keep size down and limit processor performance requirements.
Check out more of our Sensors Converge 2023 coverage.
