This question is in regard to the thresholds used as to when Sense identifies a device. Does the algorithm first identify repeated patterns in power use and then compare to the device library database to provide a device label, or is it searching the home’s power use profile for patterns from the library and if a repeated match occurs, then the device is labeled?
The reason I ask is that some devices have very distinct signatures that are easy to distinguish from other loads and can be easily identified by the homeowner. If these were identified as unknowns by the app with characteristics supplied, the homeowner could then provide the labels.
An example is a 5 hp air compressor in my shop. The unit runs off 240 Vac, The startup current is 115 Arms for about 1/2 second followed by a run current of 21.2 Arms The total time to refill the tank is about 85 seconds. That translates to 24 kW startup power and 4.9 kW run power, unlike any other load in my home. This should be easy to pick out by the algorithm and easy to identify by the home owner. Scope plots are shown below.
I don’t work for Sense, but I’m pretty sure that native detection works with changes in the patterns in the frequency domain, rather than looking at time domain patterns like you are viewing on your scope.
More thoughts on operation in the link below including some illustrations from Sense technical talks that highlight some of the parameters that are used to detect on and off transitions. Parameters like the change in RMS power on each leg during a short (maybe 1/2 second) sampling window, or the corresponding phase changes on each leg.
Please read and feel free to comment and question.
To me, the implied pattern used in native detection is in the time domain because time (short 1/2 second transitions) is referenced as a criteria. So in the case of a compressor motor used to recharge an air tank, the transitions recognized would be the 24 kW increase in power for motor starting, the 19 kW power decrease to 4.9 kW continuous power while the motor/pump is filling the tank, and then the 4.9 kW decrease in power when the motor turns off. To top it off, the fact that the power is used from 240 volts (current is equal and opposite in the L1 and L2) is simple to detect. The plot below shows the compressor turning on with the large startup current followed by about 1.5 minutes of run time.below Sense did detect the transitions, but the power deltas displayed are not the correct values. It may be that the compressor has not yet been identified as a device because it has only run 8 times since I install the power monitor. Not sure how many times the algorithms are set for identification, but standout cases like this could require less hits due to their obvious signatures.
@tktom, great to see someone scoping out their device behavior and comparing. I would love to help you resolve the three images you provided (AC Voltage, AC current, Sense Power Meter). A few questions / thoughts:
I’m assuming you are using a shunt resistor of about 100 ohms to measure the current since your measure is in Volts ?
Is you scope set up to do the measurements on the left in true Volts or RMS Volts ? If you’re measuring in Volts, then your are looking at peak-to-peak, not RMS measurements. You’ll need to do an RMS conversion to match Sense. (ps: you are saying Arms, but it looks like your measurement lines are on peak-to-peak, and the 279V on Voltage measure seems a bit of a weird number). Maybe you can walk me through how you derived your power from the scope numbers, since I’m not familiar with your scope.
The tags in the Sense Power meter won’t always reflect the change in the waveform. They were designed to emulate some of the things Sense is identifying as a waveform of interest. Given the actual Y value of the peak, I suspect that Sense was only looking at a single leg transition. The actually spike looks closer to double that, almost 16kW if you look at the scale. You can also use the measurement line on the right to get a more accurate read.
If I rough out the numbers, assuming you have been measuring peak-to-peak, then you are really seeing about 17kW - not so far off from the Sense measurement.
Once again, I’ll make the case that Sense is using frequency domain measurements because RMS is indeed a frequency domain measurement, as is phase angle. But there is a time element since Sense looks at rapid changes in those values.
Measurements were first made with a true RMS multimeter, then captured on a scope for startup timing. Multimeter values were used to “calibrate” the scope amplitudes. Current measured with a clamp on probe @ 0.1 A/mV. So my power calculations are correct at 24 kW motor startup (115 Arms x 208 Vrms) and 4.9 kW run (21 Arms x 231 Vrms). This is consistent with the label motor current rating of 21 A and the 5 hp output rating of the compressor (3.7 kW output power plus losses in the motor and compressor). That being said, the point of my post is not so much to emphasize accuracy but to highlight the signal attributes that should make it easy for the Sense algorithms to identify this load from typical loads, even if it doesn’t have my Quincy QT-54 compressor in the library. I certainly hope that the 240 V power use is included in the algorithm because in a typical home. there are generally few load of that kind.
Thanks for sharing the setup for the scope. Sounds like you are:
Using the multimeter to do the RMS calculation (I’m guessing that Sense RMS and multimeter RMS won’t completely correlate due to different RMS calculation windows).
Calculating Apparent Power, not Real Power, given the the phase difference caused by the inductive effects of the motor, especially on startup. Sense calculates Real Power (RMS).
I’m still trying to square your readings in the Sense Power Meter with your scope,
If it puts your mind at ease, Sense does have a way of detecting 240V signature or detecting two 120V signatures that closely match and merging them. Both of my AC using are 240V, in the 4kW neighborhood and Sense has found both of them.