Parameters, limits, and possibilities of Sense-ing


Sense is opening a new field of electrical forensics, which is exciting, and I’ve been pondering what types of data could be collected that might lead to further improvements/refinements - especially as some devices e.g. LED lights with many simlar/identical installations in a house may be difficult to ever detect accurately.

I’m willing to bet there are some surprising possibilities ahead; after all scientists can now pull up a core of mud from the bottom of a pond and - through a combination of pollen microscopy, geological analysis, and radioisotope dating - tell us what plants were growing there at what times in the past, when major volcanic eruptions occurred far away, and even how the composition of the atmosphere has changed over time.

So…in terms of data streams available, we have, from the CTs:

  1. Linear time domain, multiple scales
    How a device uses power over time, from inrush current through cycles of device components; the heart of Sense at present, with dramatic potential for improvement as machine learning improves and more devices in more households can be compared.

  2. Cyclical time domain
    How a device’s usage change throughout a full 1/60th sec. cycle of AC power. Power factor, and potentially even more refined detections. I assume that dimmers and transformers could be detected/differentiated in this way. I assume (though I can’t say for sure) that the current generation of Sense is collecting this type of data and using it to aid device detection

  3. Separate phase information
    Sense appears to be using this, but judging from the fact that occasionally Sense only picks up one phase of a 240V device, it seems that accuracy could be improved by defining a device as 120V only or 240V only - and this could be a user input.

Then we have external input:

  1. User input is one such, and I would like to see a somewhat larger user role in device detection, especially when trying to key in one device (i.e. a high-wattage space heater) that the algorithms are having trouble detecting.

  2. Automated input, e.g. network integration, to know when “smart”, connected devices are turned on and off - already incorporated into Sense but not helpful in my mostly-dumb house.

Then we have new and untapped possibilities. One that comes to mind this morning is active device sensing, which could work as follows:
–Use a powerful capacitor or semiconductor circuit to insert a very brief (<1 nanosecond) voltage pulse into one phase, maybe on the order of 5 volts and at different points in the AC phase cycle in order to correct for power factor.
–Using a very sensitive and fast-reading internal voltmeter, measure the return pulses in the other phase and in the neutral wire.
–Every device that is using power should return a pulse, with:
–phase (neutral vs. hot) corresponding to 120v or 240v usage
–time delay (in nanoseconds or fractions thereof) corresponding to the round trip circuit distance to each device, and
–magnitude (in microvolts or millivolts) corresponding to device resistance, and thereby to power consumption in watts

It seems like such an active-sensing technology - if technologically feasible - could be integrated into the current Sense design (perhaps with a bigger box) and - combined with the present technology - could reliably and individually detect everything using power down to the smallest nightlights and GFCI outlets. Now, if that meant that Sense cost $1500, then it might not be worth it…

closed #2