Always-on pump gets turned off


I have a 24-7 koi pond pump that I occasionally cycle when I clean the filters. This cycling was enough for it to get detected and I appropriately labeled it as a pump.

The issue is that I need to cycle the power to get it detected, which happens reliably, but then it always get turned back “off” after 50 minutes when it’s been running continuously.

Cycling the power again will get it detected for another 50 minutes…

Is this still part of the learning process?


Use the device page and go to the pump (while its running
but its not showing). Open the pump page, click upper right
corner next to cog to get to “manage”. Once on the “manage”
page, Scroll down to "report a problem and click The middle option
it’s “this device is on”.
This has helped when I have had some sporadic reporting.


If the pump runs 24 hours, I don’t think going to display because it is going to get captured into the Always On bubble. Yes, it has detected it and shows it when it first turns on, but I believe that Sense times out devices after a certain amount of time if it hasn’t seen them.

In my understanding of how Sense tracks devices - something like a motor it detects by a combination of the startup signature and then change in overall wattage. Then it turns it off when it sees that wattage go away.

In your case, it sees the start up, but eventually gives up because it is expecting it to turn back off.

I agree with @samwooly1 it can’t hurt to reach out to Support to see if they can do anything, but at the moment, my hunch is that they can’t as it means re-working how Always On calculates data. I believe they are discussing Always On, so maybe it is something they can do in the future in combination with more device specific settings like “This is a long running device” to alter how timeout’s work.


It feels like it’s timing-out.

I submitted a request and will see what comes back.


That only works for “device is not on”, it doesn’t work the other way around. :wink:

OP, if this does end up being a time out issue or the pump getting lumped into “always on” (since, in effect, it is) you could always consider putting it on an HS110 smartplug which would permanent break it out into its own device.

If that’s worth it to you or not in the grand scheme of things is up to you. If you know it’s consumption on an hourly basis then it’s per-day kWh is fairly easy to figure out from there, so having it left as “always on” isn’t that bad. Although I do understand where you’re coming from.


Your idea makes “sense” for my network rack, which is also on 24/7 and I’ve installed an HS-110 there.
It detects the pump same as it detects my Samsung TV, and if my son is up all night killing zombies that TV stays on until if gets turned off, if it ever does.
Just seems that if it reliably detects it turning on, it would detect it turning off. Maybe there’s just more learning to do?


I am speaking from a totally uneducated position, but from my overly simple view on how I think Sense works for at least some devices -
When a device first turns on, it has a very unique power spike.
So the Sense starts to see a common spike. It goes “hey, that is probably a thing turning on” After some time, it notices that every time it sees that same looking spike, about 1/10 second later, there is always an increase of 400w from just before the spike happens.
After seeing that enough time, it realizes that the 400w increase just after that spike probably has to do with it, so it names that spike and following wattage “Device 1”
So it now knows what Device 1. It knows what the power on looks like and how much power to expect after that spike its familiar with, so it feel safe saying “I found a device”.

Now 20 minutes after that spike, 400w disappears from the power usage. Because it knew that the detected device used about 400w, it is probably safely assuming that the device is now off.

The problem is, many devices don’t use an exact constant current all the time, so they have to include some fuzzy logic of some sort to allow it to deal with these things.

Lets repeat the above process, but this time, it learns that after the spike, it expects 300w and then + or - 60w. It builds a model on that.

Now you’ve got a couple of these things all going at once, and several of them have an allowance for a variable wattage. Add on top devices that unknown and things start to become a mess.

The Sense has a really good signal (the spike) to when a device turns on, but then has to rely on wattage disappearing to know make a guess if something is off.

So in comes the timeouts to make sure things aren’t miscalculated. At some point in time, it must do some math and go “have I missed an off” and then times things out.

I have no idea if the timeouts vary by device type. So they know that fridges generally run for X time, so the timeout is based on the average fridge. Same for pumps. most pumps (septic, heating etc…) only run for Y time, so they set a timeout that is appropriate. for pumps.

In any case, that is my really long way of saying I think being sure about a device turning off is harder than it seems.


I came to a similar conclusion based on several of my detected devices that I know pretty well: Sense got good at recognizing when they came on, but struggles with the “off” transition, so the total power consumption associated with these devices is pretty far off.

Note this is a bit off topic, and maybe deserves its own thread.


It’s literally on a 50 minute timer. When it goes “off”, there’s not a single ripple in my trend.

I also have a spa circulation pump that will correctly show up to 5 hours of continuous run time.

I totally agree that sensing “off” is much more difficult, no in-rush current, capacitance/inductance, spool-up, etc.