Go smarter with Edge Impulse

Go smarter with Edge Impulse

The primary constraint of RAM-1 solution is low-power design for long lasting deployment in the field which are expected to last up to 20 years. The power consumption optimization focused approach focused the implementation on using the latest in NB-IoT and LTE-M connectivity implemented as part of 4G and 5G networks as well as LoRaWAN connectivity to enable deployments in most remote locations.

Let’s explore the power related added value for RAM-1, where 25% of energy is allocated for standby operation, 50% for communication and 25% for data acquisition and creating added value. This calculation assumes however that the device only sends about 24MB of data in its entire battery lifetime, while acquiring upwards of 720MB of data internally.

This is where machine learning on the edge excels, where acquired data is not thrown away anymore by simply using it in averages or numerical algorithms, but is used to detect fine details and events from data that would be ignored. The power consumption of the actual ML model is orders of magnitude lower than actual communication.

As shown above, the raw data acquisition and numerical values can computationally provide the leakage current value from the waveform, however completely disregards any superimposed events on the waveforms. This is particularly interesting for anomalies, switching events and other sources traveling down the electrical conductor.

Izoelektro has entered into a partnership with Edge Impulse, IRNAS and Arm which came together to develop and deploy embedded machine learning models to help us build one of the world’s most advanced power grid monitoring systems with RAM-1. Leveraging Nordic Semiconductor’s nRF9160 with a Cortex-M33 processor, we created models that can run inferencing on the pole-mounted devices so efficiently, they will last 20 years on a single battery charge. This enables power grid operators to detect a range of potential faults remotely and take immediate corrective action in the event of failures.

Edge Impulse machine learning methodology enabled a rapid roll-out of ML model based on the large volume of data collected by RAM-1 product in the trial operation over the past year. We will continue to use Edge Impulse advanced solutions.

Check out a short video that we have created with our partners.

Tags: , , , , ,