Can't believe device detection can't find certain things

Well it finds things that are 50 watts, but it hasn’t found my hot air furnace that runs 10 times a day and has a distinct signature:

It starts the combustion fan and lights the flame. When the plenum is up to temp it turns the fan on. When it stops calling for heat the flame goes out and finally when it has cooled down the fan turns off. Sounds like a normal signature for a hot air furnace.

Also it hasn’t found the Keurig. This periodically turns on to keep the water hot. Again a nice repeating pattern.

I went ito this knowing device detection was mediocre and I understand how difficult it would be to detect many devices, but these two examples should be a no-brainer

Question: When your are looking at the on/off cycle in thePower Meter live on the Phone or Tablet App (not the Web App), do your see these power number “tags” being appended to those waveforms in question ?

Those tags are an indicator that Sense has a chance of detection that transition. If you don’t see them, there’s a pretty good chance your device won’t be detected.

ps: How long has your Sense been installed ? Some detections come quickly (within first week). Some can take a month or two. And some will pop up whenever as Sense deploys new models.

How can it not detect a 1000 watt change. Since I don’t see those change indicators are you saying it will probably never detect my furnace?

A few questions, first:

  • Are you looking at the Power Meter in the phone/tablet app ? The web app doesn’t show the tagging.
  • Are you looking at the Power Meter as the waveform is being measured ? Sense doesn’t backfill tags. You only see them generated as the transitions are measured.
  • Is the 1000W transition fast (within one second) or slow (more than 1 second) ?

Feel free to post a Power Meter waveform from the phone/tablet app just after it has been measured by Sense.

I prefer to look at it on a laptop, but I guess that is out of the question. I got a complete cycle of the furnace and will copy that below. I now see that the power fluctuations are there. The blower motor on the furnace is a DC motor and ramps up and down. The 50 watt ‘fuzz’ on the diagram is a heating pad and is not used at night.

That’s what I would have expected to see. You might even be able to zoom in to see more tags in the interstitial regions. I’m especially interested in whether there are tags associated with the big ramp. One of the complicated problems your picture shows, is the challenge of identifying a large slow ramp overlaid by multiple faster up and down spikes ! Shows the need for different transition identification windows.

Sense is able to detect some slow ramps, mostly for EVs and some kinds of heat pumps/mini splits. The rightmost two ramps in my image above are Model S and Model 3 charge cycles respectively. Sense picked both of those up.

The heating pad is off at night. I might be able to get a better picture soon. Of course since I have to do it on the phone it is more problematic to get a photo.

1 Like

I managed to get a cleaner shot last evening. The heating pad was not on, but at then end of the trace, you can see where the water pump turns on. It is a very repeatable pattern. Yes, the variable speed fan motor ramps up and down.

The new image looks like it is higher resolution, but it still looks like there are ton of transitions lurking in between. How long is the time axis in your screen shot ?

I was hoping you could recapture around the same 1000W change, but maybe zoom in closer ?The Sense app lets you zoom way in and way out in the Power Meter so you can see individual 1/2 second intervals or way out so you can see an entire year without missing a peak (the web app is a littler more tweaky).

Then the software seems to screw me over. Since I can only get that on my phone and not the laptop. My android doesn’t want to take a screen shot when in landscape mode and running sense software. If I use other software I can take a screen shot in landscape mode. I’ll take a picture with my camera. I’ll have to wait until later or tomorrow morning after the heating pad is off. Thanks

1 Like

Sorry it’s painful. I tend to use the phone/tablet app for screenshots because the general resolution is better and it’s actually easier than on my laptops. Thanks for your persistence.

OK, I did another short cycle this morning and photographed my phone. I then went back and annotated with absolute power numbers. So I guess it is not a 1000w change but is more like 500. Still should be detectable.

I have another furnace in my shop. Also a high efficiency job. I ran that through the same process except it didn’t turn off quickly and you can see it started another cycle. I don’t understand the ‘hump’ after the burner fires and before the fan starts, but it is repeatable. This one does not have a variable speed blower.

Thanks for the pictures. You can see the challenge here - Sense isn’t really “seeing” the whole on ramp as a single ramp of 500W or so, but rather a set of on transitions. That indicates electronics are controlling the ramp, which means that Sense can’t “read” a simple physics origin (resistive, inductive, capacitive) for the device. The off is being fully spotted in the top photo, at a whole -490W, but that means there are no directly matching ons for this device.

Sense can find some devices with these characteristics (EVs have a similar look), even though they can’t be recognized with the broad range of Sense’s more general models. But I’m pretty sure that Sense is focusing on the devices consuming the most energy that are found most broadly in their customer base, since these models are more customized.

@Shelkol @kevin1 's response above is accurate. The ramp periods you’re seeing are seen as much smaller to Sense - we’re looking at 1/2 second transitions, so Sense is really seeing small incremental jumps.

We filmed an update with our Data Science team earlier this year that highlights some work we’ve already begun on something called progressive device detection, which you will start hearing more about in the new year.

1 Like