Anybody have a electrical Kiln being monitored?

Wondering if anybody is monitoring a home with an electrical kiln in it? My wife is a potter and I purchased Sense mainly to capture her Kilns usage and costs. They use beaucoup amount of energy - like a heater using 8,500 kw every time it comes on and runs for 12 hours or more almost continuously

I have had numerous exchanges with tech support on trying to get this device correct. So far it does a great job of getting confused with my Dryer that uses 4,500 kw ever time it comes on and runs for 20 minutes. For some reason, they are not able to get it right.

I will admit my patience is getting thin with this. I have read many negative reports on device detection and don’t want to fall into that camp. I want to see this company be successful and am trying to be helpful.

However, if they can’t tell the difference between a kiln and a dryer it is like not being able to tell the difference between a hot plate and a electrical stove. Has anybody had a kiln device detected yet? Correctly?

There has been a similar line of discussion on this under EVs. Up until now, Sense has focused on recognizing on and off signatures of devices which last maybe 1/2 second at the most, because those are the kind of devices Sense can recognize instantly. Things like kilns and EV chargers have signatures that are readily identifiable to humans, but only after they have run for anywhere from a few seconds to a few minutes, well beyond the narrow on/off time window for recognition that Sense had been surveilling. Sense have indicated that they are investing in learning much longer signatures, with the idea that these are not instantly identified (bubble appears as soon as device goes on and disappears immediately on off), but are rather accounted for after the fact.

Take a look at Brad’s post near the end of this thread:

I guess I’m not really understanding this particular thread.

One of the earliest devices our original Sense detected was the GeoThermal heat pump, which has relatively long run times (never less than 20 minutes or so and often for hours during very cold weather) and a quite complex start-up process involving a sequence of deep well pump, multiple circulator pumps, the heat pump internal multi-step ramp up itself, etc. Sense never discovered many of these as individual devices, despite the fact that things like the well pump, radiant, and mini-duct pumps also operate completely independently of the heat pump itself and for much shorter…or longer….durations.

Likewise, the recent replacement Sense found the geothermal right after the Keurig and the microwave, in the first week. If Sense is only looking at short signatures, that seems puzzling. I suspect there is lots more to the story about what they look for and how they utilize that data that we aren’t aware of…one could argue “trade secrets”.

I’m just reading between the lines based on Brad’s response and the conference call Sense had a while back, buttressed a little by my personal experience. The Sense machine learning environment looks at hundreds of features embedded in the power data in order to train on and identify devices. Those chosen features plus the classification networks associated with them are core trade secrets for Sense. It’s quite understandable that Sense could spot a motor or a heat component like your GeoThermal heat pump using a half second, or shorter, turn-on and turn-off signature/features, without needing to resort to looking at the complete run time. But something that has a slower power up ramp like a car charger or presumably a kiln, requires a different set of features/models.

Yes, @andy, I reread @kevin1’s reply a number of times and feel I must be missing something. From what I have seen and learned about how the Sense system works, I think they do a longer look than just a few milliseconds. However, it may that a kiln cycle of 12 hours may be giving them something they have less experience with.

Hi @Fred_W,
May be you are misinterpreting my thoughts. I suggested that Sense uses feature window of 1/2 second for on signatures and off signatures, which is a 500 millisecond timeframe, not just a few milliseconds. Brad indicates a slightly longer window:

“When we think about furnaces and coffee machines, they have fairly distinct “on” and “off” transients which occur across just a few seconds.”

That’s a lot of current and voltage data given the Sense samples at roughly 1 microsecond intervals based on the 4 Msamples/sec at 14bit ADC/sample.

But things like car chargers take several minutes to ramp and attenuate to their on/off power outputs, making identification in a several second window nigh impossible. I think Sense hacked together a special model for the Tesla Model S since mine is identified reliably, even though it’s ramp lasts about one minute with a number of discrete steps.

Just to be clear - Sense can see the whole 12 hour waveform, but much of their early identification was based on signatures that were only a few seconds long. If you look at this blog, we’re only talking about waveform periods of only 10 or so AC cycle or 1/6th of a second…

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.