We came home from vacation Sunday afternoon to a 100-degree heat wave. Although the air-conditioner was working well when we returned home, the house was hot Monday morning. Sense showed “other” rather than the Air conditioner running, and the condensate pump was not running. The fan in the compressor unit was running, the compressor unit was making noise, but it seems the compressor itself was not running. The vapor line of the compressor unit was not cold. I called for service, the technician arrived Tuesday night and I watched as he diagnosed the problem. Noticing how he used temperature and pressure readings to diagnose the unit gave me an idea. What if these key operating parameters could be monitored continuously?
A brief Internet search identified Alert Labs as active in this area. They offer an air-conditioning monitoring product and service called Sentree. See: AC Monitoring System - Smart HVAC | Sentree - Alert Labs I sent them an email asking if they have something more tailored to homeowners and integrated with Sense. They have not yet answered my query.
A simple system would monitor the temperature of the vapor and liquid connections at the compressor unit or the evaporator unit. A more complete system would also monitor pressure at those points and monitor ambient temperature, room temperature, and perhaps plenum temperatures. This could be integrated with sense monitoring to report electric usage per degree day, or per BTU of cooling. Such monitoring could also report changes that indicate early warning of refrigerant loss, or system efficiency reductions.
If anyone knows of such a system, or is developing such a system, please let us know.
I can’t help with your primary inquiry, but you may have caught this thread where I mentioned we’re digging into data of failed ACs to help with possible failure prediction in the future. Would you be OK with us looking at your data to see if we can spot the actual failure event?
Regarding your other query … interesting! I see they were acquired by a much larger maintenance group.
More on the commercial side, I have used Monnit’s sensors to achieve similar things (leak detection; over-pressure; over/under-temp; high humidity etc … along with over-current; vibration and so on). They have been around a while and their products are quite “mature” at this point. They have saved my clients a lot, especially when you account for reduced insurance premiums. Sense can now directly manage some of that (over-current for example). Integration with Sense is still a kludge but there are good redundancy reasons to keep the systems entirely discrete (again, more on the commercial side, so I’m not really answering your question).
As far as I’m aware, btw, certain manufacturers of larger systems (Daikin, Mitsubishi [CityMulti] ) have directly accessible sensors that, if one “subscribes”, can be used for maintenance alerting. It’s easy to imagine that with the ubiquity and lowering costs of IoT & sensors that this will filter down to even the smallest systems … and fairly quickly.
The failure occurred late July 21 or early July 22. He replaced the dual run capacitor (the likely problem) and also added 3 pounds of refrigerant (not clear if this was necessary) after 7 pm on July 23.
Hi Ryan,
I don’t know if your data science team wants to add another AC compressor death to the mix, but my downstairs compressor burned out (melted plug into going into the actual compressor unit) somewhere between June 11th and June 14th. My new (April 6th) Sense unit never detected the actual compressor before it failed, but the smartplug on my furnace gives some telltale signs of the failure. I think the AC compressor controller communication with the furnace controller blower fan supersedes the continuous cooling my downstairs Ecobee is calling for.
That sentree looks similar to a product I’ve seen in commercial space to monitor ultra low temp freezers, known as Traxx. It uses unit CT’s to monitor compressor amps, what for short cycling, and with that data uses predictive analysis to assess the condition of a freezer (compressor), instead of using actual pressure and refrigerant line temp sensors which could be problematic, not to mention expensive and not reliable over time. https://www.traxxekg.com/
My AC compressor’s capacitor died last summer, and the AC unit would try to turn on every 5 minutes or so, and eventually succeed, thus I never really noticed the home being warm. I discovered the issue via Sense power meter, as it clearly showed 25kW!!! of power draw for 3-4 seconds when the AC motor would not start up on each attempt.
This sounds crazy dangerous that 100A can flow through the AC compressor, the AC disconnect/fuse (not sure what its rated at), and the 60A AC breaker…for 3-4 seconds…thank goodness the AC compressor is designed to detect such cases and cycle again in ~5 minutes. That goodness for Sense power meter to help me find the issue within a week of it starting. (Thank goodness too for the fix being just a $10 capacitor, and not an entire new AC unit, nor new home to rebuild after catastrophic fire.)
It would be more awesome if Sense could have sent an alert to me, that it had detected a possibly failing capacitor in the AC compressor.
In a way Sense can alert you to this…at least I thought it could
I believe I’d set a limit and an alert would trigger when that was exceeded but I’m no longer seeing that or know where it is. The notifications area shows the following but other than Off/On it’s not configurable…so I’m not sure where I saw it previously.
I’m almost certain I set an alert to be advised when the instantaneous usage was > 20,000 watts. “New Peak” might catch it if you’ve never had a reading that high…
I’ve had legitimate consumption of over 20kW however…all 3 AC units on, plus car charging at same time. Granted, this is very rare, and we rarely do car charging outside of off-peak prices; so having configurable peak detection notifications would be useful. (That said, Sense having AI to detect “Hey, you better check you AC system, as we detect a possible bad compressor capacitor” would be really awesome.)
I’d want “peak” detection and notification to have the ability to set both a peak current or power and a duration over which the load needs to occur. I’d want the ability to specify numerous such peak events, with different durations, and with the capability of applying them on individual devices (though I’m not sure Sense would even associate a new peak event to a known device, as the new peak current signature would look different/new).
The “goal” feature is not a substitute for peak-power (at different averaged time-scales). The “at any point” option does not have a specification for the duration; does “any point” mean any individual microsecond sample (or whatever the minimum sample rate is), or some other time scale??? The other options {end of month, end of week, end of day} are longer than what I’m looking for.
For my usecases, I would want to have peak detection/notification for:
~5s: to detect inductive motor failures.
15m: for allow mitigating of "demand charges" before they wreck a monthly bill.
Thanks for sharing @james_reilley. I would encourage you to log that request in the Product Wishlist subforum that Kevin linked to. Let’s see what others think.
Totally agree with this direction though I wonder if a better approach on this is to encourage Sense to refine the backend processing on such potential alert conditions and make explicit the existence of those refinements. I would ask: If Sense were able to alert for a bad-cap condition, would you not want the alert?
There are conflicting approaches which have pros and cons:
“I want my USAGE for ANY X-SECOND PERIOD to be under Y-WATTS” I think probably adds non-trivial complexity to UI and backend processing but allows the user base to iteratively target possible alert conditions that can feed back data to the dev team to target potential built-in, default, alerts.
“Alert for equipment failures” simplifies the front end, still allows for feed back from users via Support and here [e.g. “I’m getting false-positive alerts!”; “Equipment failure alert saved me thou$and$!”] … that could also be implemented with something similar to “Device is not on”.
Something also tells me that the period is less important than what’s actually going on during that period … the RMS wattage (Y, above) only gets you so far.
I’ve tried to create a new wishlist item for this twice this morning, but each time I just get “saving” with a spinning indicator next to it when I click “reply” to post the message. Not sure if the problem is on my end or Sense’s end.
I copied what I was trying to post this second attempt. Here it is:
I would like the ability to get a notification if peak detected power exceeds a set threshold for a set period of time. The “Goals” feature allows this for {at a moment, day, week, month} but does not provide fine tuning of the duration.
Fine tuning of the duration would allow notifications to be created for cases like:
Inductive motor startup fails, and essentially a short circuit exists through its windings, as no back-EMF opposes the current flow; 1-5s timescale.
Mitigate a higher "demand charge" before it occurs; 4-15m timescale.
I recognize that for the first case above, likely a high-pass filtering mechanism would also be desired, so that longer timescale loads can be filtered out so that their contribution is not considered in the threshold comparison.
I think this is a great idea in principle but I highly doubt that a failing capacitor is going to be identified as such by Sense which has trouble identifying the device in the first place.
I have over a year of data and my 3 HVAC systems are still not reliably detected. How would Sense identify an issue in this scenario?
My Monthly peak is typically 20000w yet one day I notice a couple of spikes to nearly 40000w. It wasn’t normal so I dug deeper and determined there was a problem with an AC unit starting up. Adding “Hard Start Kits” to all 3 units resolved the problem and the peaks were once again 20000w.
Adding the Goal as shown above “to be under” 20000k would more than likely trap this the next time it happens.
The progress of detection with Sense, that I would argue is in a generally improving direction, would seem to indicate that autonomous failure detection is at least possible. My question is not whether human input will help (it must, if only because the only way Sense will ultimately know of failure is through confirmation by the user!) but the nature of that input.
You are certainly right in that problematic device detection would complicate failure detection but in many regards the failure signature lives separately from the device. I can imagine an autonomous “bad cap” failure signature being detected without Sense having otherwise detected or tracked an HVAC system.
A big part of the equation is, of course, amassing big data … so until Sense has enough failure signatures to constitute recognizable patterns the detection method is moot. Both attacks at the problem seem valid.
