Are tankless resistive hot water heaters ever detected?

I have had sense installed for about 6 months. The device has detected many loads, but not my ecosmart 27kw hot water heater. This is the single largest user by both power and energy in the house. Really wish I could quantity how much a shower or bath costs.

Does anyone else have experience with an ecosmart hot water heater and sense?

Looking at one second data,I can see how it is a tough load to detect. Depending on flow rate and incoming water temp the load is anywhere between 2 kw and 20+ kw.

Looking at millisecond data, it should be pretty easy to identify. The heater has three 9 kw heaters and in any half wave cycle each one is either on the entire cycle or off the entire cycle. They are almost never continuously on, my water inlet just isn’t cold enough here in Southern Arizona. The peak load I have seen is about 22 kW. The heater essentially uses a form of pwm to regulate water temp, just the minimum on time is 1/120 of a second.

1 Like

Can you post a screenshot of the Power Meter showing the water heater in PCM action ? Your assumption that Sense should be able to detect half cycle (assuming you mean half 60Hz) usage is well founded, but that’s not how Sense currently leverages the microsecond level sampling. Based on what I know, the main Sense mechanism searches for significant power upticks and downticks within a half second or so window, and then uses some complex DSP analysis to pull 20 or so distinguishing / features out of the hundred of thousands of samples during that window.

This is an example of a 4 on 3 off pattern. The trace is current to one of the three resistive heating elements. RMS current is about 32 amps. All three elements are driven separately, for example: heater one might be running 4-3, heater two running 3-4 and heater three off completely to give 1/3rd (9 kW) heat. As near as I can tell, the PWM changes about once per second.

Maybe I should capture some aggregate load waveforms.

1 Like

Pretty cool that you brought out the oscilloscope ! I was suggesting to look at main Power Meter as well so you could see what a 4-on-3-off to 5-on-2-off (or whatever) transition looks like via a slightly longer than half-second analysis window (likely a 9% uptick in power). These waveforms are driven by things beyond basics physics, so Sense wont’t be able to pull out features like change in phase angle when the changes happen.

Both of your electrical knowledge is above my pay grade. If someone said get an oscilloscope I would be like, “yeah … no”.

But in my wheelhouse is the AI and the Water Heater aspect. Without getting too deep into it, electric on demand water heaters are not very common. Being that you said you’re in a warmer climate it’s tolerable but in most of the US these units pull an AVERAGE of 120 amps. Being that the majority of households only have 200 amp services … I think you get it.

See (3x 40AMP):
image

That means that the percentage of Sense users who have them is probably close to, if at all, 1%. AI is a pure numbers game. The more data you have the more likely you are to determine what something is. When your data set is very limited the learning algorithms just sit back and wait for more data before they make a judgement, or in this case an assumption of the load type. You need lots of data when applying machine learning algorithms. Often, you need more data than you may reasonably require in classical statistics.

Without much knowledge of the type of AI/ML/DL that Sense uses I would assume the following:
Basic - They need thousands of examples.
Ideally - They need tens or hundreds of thousands for “average” modeling problems.
Ideally for DL - Millions or tens-of-millions for “hard” problems like those tackled by deep learning.

So when we have devices in our homes that are more unique, they are going to be harder for Sense to find.

I originally put a poll but removed it because people may think “insta-hots” are considered on demand whole home water heaters.