How does Sense actually measure devices


#1

I am curious to better understand how Sense knows how much power a device is using. I understand that Sense has a good measure from the CTs as to how much power is actually being drawn, but how does it figure out what a given device (say a motor) is using at any given time. What seems to happen is that Sense learns the pattern a motor makes when it turns on, first the spike as it kicks on then it’s stable run wattage. Let’s say that stable run wattage is 200 watts, how is Sense able to show us a graph in real-time for that motor that is anything other than 200 watts? How does it see that as a load is increasing on the motor that the increased draw is from that motor as opposed to say a light coming on? Or does it not really see this and somehow is just bouncing around due to noise? Does it really see the increased draw or what?


#2

Great question @Howard! This gets into the nitty-gritty of how our data science team spends their time. I passed your question on to them. Here is a snippet of what they shared: “Using a combination of unsupervised machine learning, educated guessing and data we train models for each appliance. The models describe the various states an appliance could be in (eg. in a dishwasher washing, rinsing, drying, etc), the duration of each state, and the wattage consumption in each state.” Of course we don’t want to get into too many specifics since it would start getting into our “secret sauce”! Hope that helps shed some light for you though!


#3

Well that makes sense but doesn’t quite address what I was asking, but that maybe because I was not clear enough. What I hear you saying is that if something like a dishwasher has seven discrete power draws based on what it is doing Sense will learn those and display them. That is different than what I was asking. What I am asking is if a motor draws 200 watts, how is it that at times I will see it bouncing around from say 200 to 205 watts. Is this just noise or can Sense actually assign slight changes in overall draw to this specific motor at that level of detail.


#4

@Howard
My guess is that after the initial turn on, the device settles down to some value, and that Sense would use that as a baseline power for that state for that particular usage instance. Anything that varied for that usage instance that was not a different state (e.g. the rinse mode of a washer), would be interpreted as unknown, as there is probably no way one could identify the small changes.

If any one has an actual computer that’s been detected (kind of doubtful I think), I bet if you started playing a game and your GPU started pulling a lot more power, that extra power would be thrown into unknown.

I know this is maybe disappointing to many, but I think just detecting turn-on and turn-off points for devices (and maybe states of operations for devices) and total power usage at any point is probably the best we can expect.


#5

I suspcect they are actually analyzing the noise signature of our devices. Every electrical device produces electrical noise, and at great enough sensitivity there is probably enough uniqueness to differentiate between different devices.

Indeed, this is how powerline control systems like X-10 work, or powerline ethernet extenders. They inject “noise” at one end and interpret it back into a useful signal on the other end.

When you factor in manufacturing variations in power supplies and motors, interference from other things such as powerline systems and other variables I’m amazed Sense seems to be identifying as much stuff as it is.

Very cool stuff!


#6

[quote=“EricE, post:5, topic:237”]
I suspcect they are actually analyzing the noise signature of our devices. Every electrical device produces electrical noise, and at great enough sensitivity there is probably enough uniqueness to differentiate between different devices.[/quote]

The question Howard posed is not how they detect a device is off or on (which is by the turn on/off noise signature), but how they assign the amount of power to that device in use.

E.g. some TV’s use an almost constant power when on, others change depending upon the brightness of the scene being displayed. For the 2nd type, unless there is a noise figure indicating the change in brightness (doubtful), and hence a change in power, Sense can only show that device’s power as an unchanging value. The rest of the power usage would show up in the unknown bubble.


closed #7