Remote Detection and Learning


First of all, let me say that what you guys are doing is amazing and I am just diverting to look at using AI instead of a more traditional method for my project. Its one of those where you wonder - is the technology ready yet? (for my specific application anyway).

Here we go:

I would like to setup the device to detect a certain vibration pattern, such as the one in your anomaly detection method, using spectral analysis, most probably with the accelerometer but may use the mic if required.

The vibration can be triggered on either side of a thick (2 inch) plastic plane but will have the same vibration signature, although based on which side of the plane the Arduino is placed, may be a bit less loud. So could be similar to your breaking glass example- where the sensor will be on either side of the frame.

However, the sound of the anolomoly (and noise) must be sampled completely remotely over wifi. The device can be initially configured but after this it is attached to a power supply and left in its enviroment.

Would it be possible, after assigning IP properties to the device to simply install and then get the the installer to trigger the anolomoly vibration in a training mode where the vibration is manually created each time there is a beep, so that the samples are captured with the correct begining.

After the training, the device is left alone and then I can just program it to send signals when the anomaly is detected.

Is it possible to use edge-impulse plus Arduino BLE sense (or any other board) in any way?

This is a highly mission-critical application, so could you also please comment on how reliable you would say the anomaly detection could be if trained properly.


Hi @j45p41, great idea, and I think this sounds like a great usecase for Edge Impulse. Reliability is hard to comment on at this point, but I think you’ll get a pretty clear idea how well this will work once you’ve collected and validated ~10 min. of data.

To do the initial data capture you can send data directly to the ingestion service with a simple API call. If you have a WiFi radio hooked up to your device you could make the device send a beep, then collect the data, then send it to the ingestion service.

We’ve done work on WiFi-enabled boards (e.g. the ST IoT Discovery Kit) and there we also implement the remote management protocol, so we can capture data from the computer while the device is connected to a machine - but this probably won’t be necessary for your usecase.