Is it common for the device detection to go backwards after a month?
After a month, my Sense had found my heat pump compressor (called AC) but not the blower. It had also found 2 water heaters. Finally, it had just found my electric car. Things were looking up.
For the past week, things are going backwards. The heat pump runs several times per day but Sense doesn’t identity the use as AC any more. It hasn’t shown usage for 9 days. The EV, although identified, hasn’t had any usage, despite charging daily. The water heaters are only showing intermittently.
The only things that seem to be detected regularly are the garage door opener, refrigerators and freezers, and the dishwasher.
Is there anything I can or should do to improve the situation?
It may not apply in your case but I know if a CT moves, even a little, it can throw things off. I rotated one of mine about 90 degrees without removing it from the wire. My Sense immediately quit seeing my refrigerator which it had tracked reliably for several weeks. My “Always on” usage also went up about 4%. I didn’t have many detections yet so I did a full reset to start over. A support person later told me that was the right thing to do as my moving of the CT threw off the calibration.
I’m having the exact same issue - no movement of CT - several devices that were working very well are now going backwards - very frustrating - contact support and all they say is to delete devices and start over
That is very troubling. Say in two or three years I have some electrical work done in my panel and in the course of the work the CT clamps get moved in some way, which is a virtual certainty as my panel is pretty packed. Will that throw off or negate years of detection necessitating me to start over?
I think that’s hogwash. Moving the ct shouldn’t cause any issues at all.
Opening and closing clamps may affect current calibration, but hasn’t in my experience affected device detection. It’s a red herring.
These issues mentioned in this thread happen to everyone, some more frequently that others.
I agree with this. The CT is a transformer split-core that when closed forms a complete core. The magnetic fields that are captured are only the wires going on the inside of the core. 1 of these is the wire it is clamped around, and the other is the output wire wrapped around the core and feeding it’s wires. Where it is located along the clamped wire should make zero difference.
I was contacted by Sense a week ago saying that there was a problem with my mains CTs, because one was not closed completely. I checked, and they were closed (and they couldn’t be pressed open after the panel cover was put on). Nevertheless, I opened and re-closed them. Then they said since I opened and closed them, I had to reset my system and recalibrate (which makes no sense to me), which I did, even though it seemed to be irrelevant. I’m not sure what they had expected me to do whether a CT was open or not, but I don’t see what difference it would make to require a reset. Once a calibration was done on a properly clamped system, the values should be whatever it reads.
If a CT is not totally closed, the values would definitely be a bit lower on that side.
I didn’t mean to imply that whenever a device goes away it’s because a CT moved. I’m guessing the most likely cause is simply how the detection algorithm works. It’s probably always relearning and at some point may see something that it likes better than its previous results. It surely does that or it would never correct errors. Anyway, I only mentioned moving a CT because doing that caused changes to my reported devices. It probably had nothing to do with the centering of the conductor. While manufacturers say keep the wire centered or at least away from the split for the best accuracy, the biggest error due to position I could find documented for this type of clamp was only 2%. So davidferri, I don’t think you have anything to worry about. Maybe mine wasn’t closed originally despite my best efforts. Or maybe I have a bad connection or who knows what. By “moving it even a little” I may have corrected a sub-par installation. Or created one. I was just trying to point out that if you mess with the CTs after the Sense has done its calibration it could cause devices to go away. For better or worse.
Since it had been more than 10 days since the devices had seen activity (even though they were still present and drawing power), I decided to reset my system and try again.
As far as I know, my meter and clamps haven’t moved. They are in the box and it is closed up, so there certainly wasn’t any intentional movement.
It has been almost a week since I contacted support about the issue and I hadn’t heard anything from them, so I decided to just reset the system and hope that it works better.
When I installed mine, it very quickly identified the furnace, and as a separate item, the furnace blower. It worked this way for about a week when it seemed to forget about the blower. Any power originally allocated to the blower was now going into the unknown category.
It worked like this for about 6 weeks until a miracle occurred (I call it a miracle because I was never expecting this). Suddenly, both the furnace and the blower were lumped into one device! Yippee!
So it took awhile, with an unlearned device, but everything eventually started working as it should.
@mike1, I’m sorry that support did not get back to you yet. I understand you have already reset your monitor so it may be a little late for them to be able to look into it. If this happens again, please reach out to support again so that they can take a deeper look at your data.
@dgreen045, I saw that Francisco, our support team member, recently responded. This issue should correct itself as Sense continues to learn your A/C device. If you continue to see issues, be sure to reach back out.
I continue to have the same problem with my two AC compressors. My downstairs compressors show no use for five days and now I do not see a new one being detected.
I’m sorry to hear that Andy. The team is working to improve device detection processes and build features that enable users to be more involved in the improvements. Recently, we released a feature that allows users to mark devices as ‘not on’, if they incorrectly appear as on. The data science team will be using this data to improve device detection. We are working on other features along this line.
@dgreen045, have you let support know? They’ll be better able to get dig in and see what’s going on.
My 2c after 4 months of watching the Sense recognize some devices well, recognize others, then begin to conflate them with different devices, while never detecting others:
This is a hard problem - some experts believe that disaggregation of devices is nigh impossible. I like that Sense has brazenly attempted to take on the problem, especially without self-guided (at least from our end) machine learning.
I’m seeing many of the same problems I saw in early facial recognition with both Apple and Google products. The learning system can recognize some subset of easily recognizable device types. But it never finds stuff outside it’s models. Plus it occasionally mis-identifies new items as an already recognized item - once that happens, recognition of the erroneous device gets reinforced/amplified and eventually is fully conflated with the originally correctly identified device.
Exactly. Monitor was making slow but steady progress. Now devices are being confused, I have ghost devices, etc. Too many issues to report here, though all seem to be shared by others in other threads. Disappointed. -MB
A week after resetting my Sense unit, the device detection is going better than the first time. I have discovered everything I had before except the second water heater and the EV. It has also detected the dryer and coffee maker that it hadn’t last time. It still isn’t reliably assigning usage to the heat pump, but I am hopeful.
I suspect the heat pump is hard for it because my well pump has to run to supply water for it. Because of that, the Sense sees the AC unit very little without the well pump mixed in. With summer approaching, I will run the well pump more without the heat pump. Hopefully, the pump will be detected. It is variable speed so it has a really unique power signature.