IoT devices are under-utilized - they have amazing sensors and powerful processors, but we use them to just send single temperature values. This is all because we are limited on power and bandwidth, the only way to reach 2, 3 or 10 years of battery life is sending as little as possible. Thus, to make smarter devices we need to process sensor signals directly on the IoT device and just send the conclusions over. Slides from my talk on TinyML and Edge Impulse at The Things Conference 2020 in Amsterdam.
3. Typical LoRaWAN sensor in 2020
Vibration sensor (up to 1,000 times per second)
Temperature sensor
NFC
Water & explosion proof
Processor capable of running >20 million
instructions per second
4. But... what does it actually do
Once an hour:
• Average motion (RMS)
• Peak motion
• Current temperature
5. 99% of sensor data is discarded due to
cost, bandwidth or power constraints
https://www.mckinsey.com/~/media/McKinsey/Business%20Functions/McKinsey%20Digital/Our%20Insights/
The%20Internet%20of%20Things%20The%20value%20of%20digitizing%20the%20physical%20world/The-Internet-of-things-
Mapping-the-value-beyond-the-hype.ashx
6. Physical world holds a lot of data
>
hZps://rakuten.today/tech-innova[on/simon-crosby-smart-ci[es-edge-compu[ng.html
9. On-device intelligence is the only solution
Vibra[on paZern
heard that lead to fault
state in a weekTemperature
varies in a way that
I've never seen
before
Machine
oscillates different
than all other
machines in the
factory
10. On-device intelligence is the only solution
Temperature varies in a way
that I've never seen before
(0x1)
26. Training vs. classification
Hundreds of different states
Need to encounter states many times
Training takes long!
Classification is however simple
Play the game, and you have to open up max. 4 drawers!
35. From model to device
Signal processing, neural network and
anomaly detec[on
36. Conclusions back to TTN
♻
Sample for four seconds
Classify
Result differs? Message.
Sheep is walking
hZps://pixabay.com/photos/sheep-curious-look-farm-animal-1822137/