Sense not recognizing identified devices

Sense recently identified my food dehydrator a week ago, but now after plugging it back in to use it again, Sense isn’t seeing it anymore. Does a device’s electrical signature slightly change if I plug it into a different outlet than the one it was discovered on? Thanks!

Yes and no.
It certainly is possible that the plug you are using now is in some way affecting the electrical signature.

Another possibility is that the plug you are using now is on the other “half” of your electrical panel. I have no proof of this other than my own experience, and I know others have not had the same experience as I, but I have found that Sense seems to look at specific sides of your panel when it is trying to recognize a device.
I have a dehumidifier in my two room basement. I can move it to any plug in one room and it is detected. If I move it to the other room, it is not detected. The outlets in the other room on the other half of the panel from which the dehumidifier was originally detected.

I have never left it long enough in the other room to see if Sense also learns it there, but my guess is that it eventually would. The question would be that if it learns it on the other half of the panel, will it learn it as a new device, or will it recognize it as the originally, but just “know” that now that device might show up on either side of the panel.

4 Likes

Addressing your question @c.swaggner and responding to @ben: I believe the difficulty with tracking a device that has changed to an outlet on a different leg (phase) of your household supply is most likely due to the phases normally being different enough in voltage that the electrical signature may vary. Sense is only logging integrated (current x voltage) = watts and this is what is used for learning devices. A voltage change due to switching which phase a device is plugged in to doesn’t necessarily result a wattage change (the current can vary to compensate) but in some cases a change in voltage will change the watts used. For example, I believe a dehydrator fan supplied with increased voltage will increase in speed and use some more watts. Or reduce voltage, reduce watts.

The other factor would be circuit resistance (for the same reason). A long supply wire to an outlet (or an extension cord!) will have higher resistance and the output voltage will be lower. Again, depending upon the device this could result in higher logged watts and throw off the Sense detection because the electrical signature will be different. e.g. a 100ft long standard #12 wire a 4A load on 120V will have a voltage drop of about 1.25V or 1%. This could easily result (for some devices) in not being recognized. As a general rule, I’m guessing the lower the wattage demand of the device the harder it will be to detect if the supply voltage or wire length changes.

As others have stated, this is most likely due to toy plugging the device into an outlet that is on the other “phase”.

However, sense wouldn’t just do detection based on the power consumed, they are looking at both voltage and current signatures. I don’t believe a 1% change in voltage should throw off the detection, after all your supply voltage varies throughout the day as well. I’m guessing that this is more due to the fact that you have to CT clamps, and they are calibrated to the particular input. This is why switching the CT clamps screws up the detection.

Thanks for all of the input, everyone. I think in the meantime I’m just going to be cheesy and use the supplied stickers to designate outlets for my devices, since I have 7 outlet panels in my kitchen, so it’s not like I’m going to be hard pressed to find a spot for something else haha

Yes, agreed that a phase switch is the most likely cause of detection issues but I stand to be corrected on a wire resistance change being problematic, especially for devices like motors and fans. I’m falling back on the voltage variation between phases being the issue along with the phase shift and mismatch with whatever magic Sense is doing for detection.

Hmmm… I’m not sure that’s true regarding using current & voltage recording … double the data for (perhaps) no great gain and lots of added cost? While it would be ideal to compare both waveforms all the time, my understanding is Sense was just doing ML on the stored aggregated waveform: i.e. power. That said, it would be nice to know that voltage & current are being stored because then we could get historical voltage readings and graphs one day!

@kevin1 has detailed some of this on this thread and here.

I just poked through a couple of the patents, Privacy docs, blogs and herein (the forum) but actually can’t find a clear statement about that one way or the other. Engineers? (@kevin1 ? @RyanAtSense )

Does Sense permanently store separate current and voltage waveforms (for) all the time? At the same frequency

[Can’t believe I don’t know this already!]

@c.swaggner, stickers rule! Sense is doing dehydrator stickers now? :smiley:

I don’t know if Sense stores historical voltage/current data, but it appears that they use both for machine learning. Here are some excepts from their homepage:

“Sense uses this high-resolution data to detect small changes in magnitude, phase and frequency, and then applies machine learning algorithms to distinguish one appliance from another — to recognize individual voices in that noisy conversation”

“As the filament heats up, it draws a lot of current, but then stabilizes. The current and voltage are in phase with each other, so Sense knows it’s a resistive load.”

Yes I’m aware of this.

The question though, partially answered by @kevin1 and referenced prior, is the use of historical data vs “instantaneous” or non-Mothership (AWS ML) assessment – determination made using instantaneous readings by the Sense monitor itself.

e.g. a resistive load, with current and voltage in phase, can be determined over a short time period by the Sense monitor and then, lets say, aggregated power data (i.e. not current and voltage, just watts) is stored in the Mothership (AWS) and tagged as “resistive load”. This, for the most part, is all you would need and would save a lot of Mothership storage and processing overhead.

Detection takes place on both lines separately. You can hear more about this from the DS team in this video

We store historic current, voltage, and frequency data. This gets utilized in the device detection and model refinement processes, as well as by the Support team in diagnostics. This, of course, doesn’t mean we intend to populate it on the user side any time soon. On the grand scheme of features that would benefit the bulk of our users, that one is pretty low.

3 Likes

You should not have to do that. Its crazy. What about a vacuum cleaner? Its not going to be recognized on one outlet but it is on another? If Sense is that sensitive to outlets, it just not that useful.

Eventually you could imagine Utilities offering some of the features that Sense has … like realtime, high frequency tracking and logging of actual overall electrical usage … but at Utility scale it’s hard to see how this would be achieved anytime soon industry-wide. More likely is that people who see immediate benefit from fine-grained energy tracking (both use, and generation = solar) will adopt monitors like Sense and extract as much benefit from them as possible. “Have it pay for itself in energy (and disaster) saving”.

Meanwhile on a micro level, Sense is certainly challenged by tracking all devices all the time with any kind of accuracy and they make that fairly clear … it’s a learning experience for the Sense itself and the people who are building it … and people like us who give them feedback.

Devices that are moved around and plugged in to different outlets are probably at the top of the list of hard-to-detect, especially ones with complex varied signatures like a vacuum cleaner. There’s nothing particularly “standard” about the way houses are wired so different phases, wire gauges, wire lengths (and coiled wires in partially retracted vacuum cords) and so on come into play in detection. That said, Sense is not unaware of this and from what I understand given the responses in this thread the pool of data is growing and detection is improving.

The more people plug in their vacuums all over their Sense-connected houses the quicker the Sense brain will learn.

2 Likes

@rgmichel

I don’t think what you’re saying necessarily follows from what the video said. Did you read @ixu’s post? It provides a more detailed explanation of how plugging into an outlet on a different phase circuit would affect device detection.

But maybe I’m misunderstanding what you’re worried about. I guess the question to the forum is, has anyone seen a device be identified properly regardless of which outlet it is plugged into?

I can’t test this myself since Sense hasn’t identified my vacuum cleaner yet. Note that this doesn’t bother me that much since I know the vacuum wouldn’t be a significant portion of my electric bill. I have to remind myself of that occasionally as I get obsessed about getting to 100% detection. It’s a disease!

1 Like

Indeed @dianecarolmark we can all get caught with OCD … Obsessively Counting Devices.

To your question: Yes, that’s a good way of putting it. I actually think I haven’t seen device detection on devices that have been moved BUT I also haven’t had any device detection on anything moveable! Meaning all my device detection has been on permanently wired devices (not necessarily hard-wired though). Mine is not necessarily a good test bed because I have a LOT of tools and chargers and electronics that would be quite noisy for ML. I have 5 vacuum cleaners for example. And if it’s any indication (it is) my electrical panel cover is more often off than on. As I’ve mentioned, the big things (hot water heater, garbage disposal) have been recognized but I turned to Smart Plugs (and a second Sense on 208/240V devices) for devices I want to track.

If its any consolation regarding your disease (it is!) you get 100% detection (well, 98-99%) for ALL devices all of the time in the form of the overall energy tracking. Eventually there will be Sense vs Human battles so you can start training now!

I am a new sense user, about 1 week. The only electric usage is the fridge, cable and WiFi box, and kitchen equipment clocks. Yesterday we turned on the septic system dosing pump. It’s electric usage is distinct. How can I identify this piece of equipment now? As we move int the house more and more equipment will go active. How can I identify the devices as they light off ? Thanks
John

That’s a somewhat open-ended question and maybe you could create a new thread with your specific expectations. What I am understanding is that you may need to (now) figure out whether the septic pump is actually working … given you just moved into the house and installed Sense?

Yes, in can be tricky given the other usage in the house to narrow down whether something is even switching on or off, let alone the actual energy consumption.

  • Is the device on a separate circuit that you can identify? (a pump could be/should be depending upon local electrical code I believe).
  • Can you identify when the device is operating through means other than Sense? e.g. Can you hear it?
  • A pump usually has a distinct electrical signature and timing … if you know the make/model you may find it talked about (or similar) in the forum and even find a similar waveform posted. Try searching for “pump”.

To preempt what might be over enthusiastic expectations: new Sense users are often confused about the detection process and what it implies when you look at a detected waveform. There are a number of explanations of this on the forum.

Rayshawn gives some good approaches for matching actual devices with detected devices at this point in the April video.

1 Like

Indeed @RayshawnAtSense’s description is great for detected devices but from what I understood of @jswiatek77’s post (and I was addressing above) is the pre-detection scenario. Not that looking through the forum as @RayshawnAtSense suggests doesn’t help to work out what, say, a pump might look like in the overall Power Meter … it certainly does help!

I think the trick for new users when there is no detection is to understand the limits. Either you have to wait for Sense to do it’s magic or start to use human intelligence beyond the scope of what Sense is displaying and has access to. Look around; flick some breakers; is it hot?; does it smell? Staring at the waveforms only get’s you so far.

2 Likes

@ixu,

All good points… If someone wants to associate Power Meter waveforms with not-yet-detected devices turning on and off, they really need to wander around the house with the Sense App up, turning things on an off, or doing things that might cause a device to turn on (lowering fridge settings, flushing toilets to reveal the septic pump).

1 Like

I can see the pump specific operation period in the power meter. It is a 30 second spike of about 1,300 watts over and above the other small load items that are always on. How can I tell sense that the chunk of usage is specifically the septic dosing pump?

If only it were that easy!

Short answer: Wait.

Long answer: Sense needs to learn how the device behaves. Just because you know it’s the pump doesn’t mean Sense has a clue yet and unfortunately even if there were a way to tell Sense what you’re seeing it wouldn’t exactly help the algorithm. There seem to be many Sense users who have pumps that have been detected so you can have some confidence it will be recognized … especially given you don’t have a lot of other devices running. That said, you may get (un)lucky and the pump will get detected quickly BUT as you add more to your house the detection may falter as the electrical circuits get noisier.

The Sense team, as far as I am aware, has debated (and perhaps continues to debate) the efficacy of teaching Sense vs letting it learn (you, like others, would think in this case “We’ll can’t I at least tell Sense I have a pump connected?”). We could go deep here but I’ll point you to their recent explanation and hope that clarifies things.