New user here hoping someone can help me understand the following. Why does the “now” chart show +3082 watts for my boiler yet the device specific page shows approximately +750 watts for the very same point in time when the boiler started?
If the top line of the “Now” chart was 3000 watts at the time this image was taken then the midline would be 1500. Right off the bat any individual device going +3082 should blow past the top line in that view at which point it would adjust to a new higher value. The total power in use at that time doesn’t show as >3000 watts.
The total power spiking a bit over 1500 when the boiler first kicks in aligns much better with the ~+750 I see on the boilers device specific page.
Thanks in advance for any ideas or helping me understand how to read this.
The “now” page says the boiler had a +3082 spike, but the visual is nowhere near 3000 watts.
Another slightly different example of numbers and visuals not adding up. Here we see a +2900 is just shorter in size than a nearby +1400. If the boiler went up 2900 it should be much closer to the top of the graph since the Y axis in this particular view was 3300. The 1400 is a Keurig coffee machine it has not yet learned. Is it a matter of the inrush spike duration being too short to have a point represented properly on the chart due to data averages over the sample period, but the chart still shows the highest value it saw?
There are a couple of possible explanations for this, but recommend you send the charts to support@sense.com for their review as the graphics you are seeing are confusing.
Explanation 1 is a software bug, most likely if you are sure the absolute peak inrush current, presumably starting a blower, is nowhere near the value displayed.
Explanation 2 is that the peak value is in fact correct (Sense samples at a very high rate to get as much frequency content as it can as part of its device recognition process) but that the sampling rate used to generate the graph results in a much lower peak value, whereas the number displayed is the actual peak.
I’ll put my 50c on the second, although I’ve just looked through some rapid transients in my data and can’t find any examples of the same problem
I’m leaning towards the second as well. I’ve had the same type issue in my own career doing server performance analysis. I may be able to see a value when looking at the raw data samples, but the visualization is skewed due to each line being an average of a couple samples to smooth out the view and reduce visual peaks &valleys.
Device history shows almost always a +750 or +1200 when this one kicks on. I may see if I can get my own current clamp over the wiring and catch what it says is the max witnessed value.
Unless you can connect a current clamp directly to a storage scope, I suspect a typical clamp meter isn’t going to get anywhere near the sampling rate Sense uses.
Let’s tag @RyanAtSense and await his input on this interesting one.
Definitely an interesting issue! I’m leaning towards @dave_n_s explanation 2, but let me talk to some of the engineers today and try to get a better idea for what’s happening.
Regardless of what’s found it would be great if the charts always accurately represent the spike it reports seeing (the +### value), even if only for one tick-width on the chart. If the app feels it is important enough to tell me about the +#### value, then it should also be important enough to accurately display it so I can scroll back and see the total wattage in use when the spike was reported.
So, it sounds like @dave_n_s was pretty spot on with his 2nd hypothesis. Essentially, that spike likely occurred at a smaller interval of time than the individual device viz shows. But go ahead and shoot me a PM with your email or monitor serial and we can find out for certain.
For consistency sake my thoughts above would equally apply to any -#### dip value. It’d be lovely if these highs and lows are more accurately displayed ‘now’ chart as well as able to be found/viewed in the device’s charts. As one may imagine it felt really weird being told my boiler usage up +3000 and then seeing nothing even remotely close in it’s individual device chart. You start to get a “which should I believe more or are they both wrong?” kind of feeling.
Very likely. Did anyone else watch Silicon Valley last night? There was a classic example of this possible scenario when they were bringing their two newly acquired dev teams together.
I’m not sure we want to see the live waveform data down to a single 60Hz cycle, though sub-cycle data is useful to Sense for some identifications. It would be nice to know what kind of filtering / windowing / smoothing are applied to the data Sense displays, both for the power meter graph and the associated markers…
Just wanted to give a quick update on this… there’s a few people looking into it and it’s a particularly weird issue, so I may not have a solid answer for a couple days. But, I assure you, I will post in here once we’ve got it figured out! I appreciate the patience.
Hah! Thank you for the honesty, Ryan. I have a tendency to stumble across odd-ball scenarios like some incredibly terrible super power I’d rather be rid of. Certainly interested in what the team finds.
I just wanted to give everyone an update on this. Engineering has confirmed that there’s an issue with scaling and it is currently being worked on. It shouldn’t be affecting most people’s day-to-day use of Sense, but if you’re noticing similar issues, don’t hesitate to either let me know or write into support@sense.com.
The device-specific page is correct. Essentially, the power meter is calculating wattages in a different way that is causing issues with how it displays the wattages for some devices, based on the half-second resolution of the Power Meter.
@RyanAtSense, have there been any updates in this area? I’m still seeing things like the “now” chart showing “Boiler +3116w” yet if you go to the boiler page it shows more like +225w. I noticed as I was turning on all of our major appliances to see where we’d end up for current wattage.
I think I saw an answer back in the Ben days that the annotation in the Power Meter was the spike in the on-signature but the data in the Device “Power Meter” was more like the steady state power over a longer half-second interval(as Ryan confirmed). But without more specifics on the spike accumulation intervals, it is definitely hard to reckon with what that number really means.
I have a tendency to stumble across odd-ball scenarios like some incredibly terrible super power I’d rather be rid of. Certainly interested in what the team finds.
