Hello everyone. I have at times when the house is quiet like first thing in the morning and usage is low tried to isolate a device so I could label it.
I have several scent machines that turn on, run for about 30 seconds or so then turn off. I unplugged all except 1, and watched the power meter as it would turn on and off. I can see the jumps in the meter and was thinking, why can’t I just click on that spike and create a device by clicking on the spike. I know what the spikes are, they correlate exactly to when the machine kicks on and off, but Sense doesn’t isolate it as a new device.
So, is there a way to make the power meter interactive where I can click a spike and ID it from there? Would seem to be pretty useful feature.
You’re asking a very good and widely asked question. The reality is that on and off spikes have to be “found” by Sense, before you can give a name to them. That’s because the on and off spikes have to meet a variety of criteria before they are recognized as unique device. Here’s how to think about the Sense process before Sense allows you to label a transition / spike.
First Sense has to be able to “see” a transition. With the current native detection, Sense looks for up and down transitions spikes that last less than a second or less. This leaves out charging cycles for most EVs and operating cycles of many types of air conditioners, because the ramps are slower.
Second, it is helpful if the transitions / spikes have physics “fingerprints”. Most simple things like motors, light bulbs, heating elements and microwave ovens reveal additional evidence about their identity in the relationship between the voltage waveform and the current waveform. Unfortunately more complex electronic devices give a much more scrambled and unreadable “fingerprint”. Sense might see the transition / spike for those devices, but be unable to “file” it in a consistent place. Plus many transitions/spikes coming from electronic devices don’t correspond to on/off, but rather the music getting louder (stereo) or a sudden change in video output (a commercial coming on an LCD TV).
If the transition / spike has a clean fingerprint that allows it to be filed away, it gets collected by Sense. As Sense collects more and more, Sense tries to find similar fingerprints that are unique. That means it needs to find a bunch that are very similar across all dimensions. They are allowed to vary a little because other things happening in the house can sometimes modify the fingerprints a little. One common problem determining uniqueness, though, is that there can be many fingerprints that are all close, but instead of forming a self-contained cluster, they form a huge band without enough delineation within the band. This can happen when you have multiple similar devices like your scent machines, all with slightly different fingerprints. Sometimes this will prevent Sense from “detecting” a single device out of that band of devices, but sometimes it will instead “detect” that broad band as a single device, resulting in device conflations.
Finally, Sense must find a matching unique off for every on transition / spike. It’s not useful to only find ons (or offs) - both paired together are needed. Once Sense finds a unique on and off transition / spike pair, as described, it will pop up and ask you to name it.
That may not sound like a very satisfying answer, but it’s the same reason that photography programs that do facial recognition must first recognize something as a face before you can add a meaningful name that the program can apply to other faces it finds.
A longer version with more technical details and pictures here:
Thank you so much for the detailed answer it actually makes a lot of sense. An interesting thing before I saw your answer I noticed in the power meter a lot of these spikes seem to have almost exactly the same profile. Like you could super impose them on top of each other and they would be the same down to little peaks and valleys at the top of the spike and are same or what I would call statistically similar in duration. Like looking at someone EKG.
My problem is going to be I have 4 of these devices two of which are different manufacturers and they all run about 30 seconds and off for like 2 minutes, you can vary the time so they are all not on or off at the same time.
I’m presuming they would all have similar draws, but I may need to group them if Sense ID’s the spikes, it’s never going to know which scent device I’m sure.
Slightly off topic but another interesting observation is the power meter will show spikes throughout the night (ie 1am , 2 am etc….) and you gotta wonder what the heck is causing spikes when the house is quiet. If it was HVAC kicking and off the duration would be much longer so there are these mystery short duration Al spikes throughout the odd hours of the night and early morning!
Spikes at night or when no one is at home does make one wonder what is going on. Many of these seemingly strange spikes are quite likely completely normal and could be anything from a refrigerator/freezer compressor, defrost cycle, ice maker, sump pump, dehumidifier, Keurig or similar coffee maker, aquarium heater, a smart TV or cable box updating itself, a charging device. Knowing the wattage and duration can help narrow them down.
Sense may or may not eventually “discover” some of them.
They could also be a component of a discovered device. For example Sense might discover what it thinks is a refrigerator. Often it’s only the compressor and not the entire fridge, as to Sense the compressor’s normal cooling operation looks quite different than the defrost cycle, the ice maker or the light.
Washing Machines can be tricky as the electrical usage and patterns for different wash cycles wash, rince, spin, not to mention different garment settings are quite different. Sense has the difficult job of trying to determine (machine learning) if those different patterns are of the same device or multiple devices.
Imagine how difficult it is to differentiate between multiples of the same device.
It’s fairly easy for someone to pick out a small device in their home on a graph when things are electrically quiet and they know what is going on at the time. When your not at home have 3-4 other people running around the house turning multiple things on an off randomly and try to pick out that same device in real time at the exact moment it turns on and off in addition to all the other devices. Sense has to essentially do that and recognize it with enough certainty and consistency to seperate, determine it is it’s own device, not part of another device or devices in order to it to be “discovered” as it’s own device. Not to mention enough other users with the same discovered device to compare and determine the device type.