Before bothering support I thought I’d ask what the experiences others have had with their AC systems.
We’re finally starting to have warm days where both our 1st floor and 2nd floor AC systems are running. Each system has its own condenser and its own air handler.
Regardless of heating/cooling season the air handlers run for 5 minutes per hour to circulate air in the house. When only the air handlers are running to circulate air Sense seems to do a good job of knowing which one is on or off.
However, when the condensers are also running a couple things take place.
Sense often says the wrong condenser is running.
The air handlers remain in an “off” state even though they’re running.
The systems are essentially the same other than their nominal capacity ratings. Model # differences highlighted below.
I have a very similar setup. 2 Bryant forced air systems with Bryant condensers. The upstairs system is sized smaller than the downstairs system, though the upstairs runs more frequently. I also have the air handler / furnace fan set to run for 5min / hr even when heating and cooling are not on. I reset my Sense back at the end of August last year to try to improve recognition - here’s what I have seen.
Sense quickly recognized both air handlers and was able to tell them apart with very good accuracy. By October it was also picking up AC condensers, though the air handler seemed to be subsumed by the condenser turn-on.
I thought everything was working nicely because I did occasional spot checks when I heard one of the furnace fans turn on. But the monthly turn-on statistics tell a different story. Without AC or a furnace cycling, we should see between 700 and 744 turn-ons per month. Sept-November look reasonable but things go a little crazy after that. The upstairs fan runs seem to begin to slip into downstairs identifications. And then detections seem to begin disappearing altogether. Maybe I’ll get brave and download monthly .csvs for both my Ecobees, then process into actual times-on per month for comparison.
Something else odd I just noticed. As I type our second floor condenser and fan pair are running. The 1st floor AC hardware is not running.
Sense is showing the 1st floor condenser as on and my 2nd floor fan as off. My 2nd floor condenser and my 1sf floor fan devices have gone missing from the device list.
Maybe it’s time to hit up support as I haven’t merged any devices.
Here’s Sense results for number of times on in a month vs. Ecobee’s definitive numbers. Looks like Sense is doing a pretty good job identifying the condensers, but has drifted away from correctly identifying the furnace fans / air handlers. Quite honestly, the Ecobee data is not entirely definitive since it has dropouts due to various WiFi / power weirdities. I’ve seen 300-1000 5min blanks per month.
BTW, here’s the R code for reading in one month of Ecobee data and returning the number of Fan, Heat, and AC times-on. Needs to be customized to remove unneeded data from your specially named room sensors. See comments.
library(readr)
# Read in Ecobeee CSV file that has been renamed Month.csv. Skip extraneous columns
# Will need to be customized to skip temperature and occupancy data for all the room sensors (RoomA, RoomB in this example)
EcoMonth.df <- read_csv("~/Downloads/Month.csv", col_types = cols(`Cool Set Temp (F)` = col_skip(), `Cool Stage 1 (sec)` = col_number(), `Current Humidity (%RH)` = col_skip(),
`Current Temp (F)` = col_skip(), `DM Offset` = col_skip(), `Fan (sec)` = col_number(), `Heat Set Temp (F)` = col_skip(), `Heat Stage 1 (sec)` = col_number(),
`RoomA (F)` = col_skip(), `RoomA2` = col_skip(), `RoomB (F)` = col_skip(), `RoomB2` = col_skip(), `Outdoor Temp (F)` = col_skip(), `System Setting` = col_skip(), `Thermostat Humidity (%RH)` = col_skip(), `Thermostat Motion` = col_skip(), `Thermostat Temperature (F)` = col_skip(), `Wind Speed (km/h)` = col_skip()),
skip = 5)
# Simplify column names
colnames (EcoMonth.df) <- c("Date", "Time", "Mode", "Event", "Program", "Cool", "Heat", "Fan")
# Create previous state columns to ease state change detection
EcoMonth.df$PrevCool <- c(0, EcoMonth.df$Cool[1:length(EcoMonth.df$Cool)-1])
EcoMonth.df$PrevHeat <- c(0, EcoMonth.df$Heat[1:length(EcoMonth.df$Heat)-1])
EcoMonth.df$PrevFan <- c(0, EcoMonth.df$Fan[1:length(EcoMonth.df$Fan)-1])
# Find the on switches where previous state was off and new state is on (time > 0)
# Need to also screen out balnks in the Ecobee data that show up as NAs after reading into R.
EcoMonth.df$FanON <- (EcoMonth.df$PrevFan == 0) & (EcoMonth.df$Fan != 0) & ! is.na(EcoMonth.df$Fan) & ! is.na(EcoMonth.df$PrevFan)
EcoMonth.df$CoolON <- (EcoMonth.df$PrevCool == 0) & (EcoMonth.df$Cool != 0) & ! is.na(EcoMonth.df$Cool) & ! is.na(EcoMonth.df$PrevCool)
EcoMonth.df$HeatON <- (EcoMonth.df$PrevHeat == 0) & (EcoMonth.df$Heat != 0) & ! is.na(EcoMonth.df$Heat) & ! is.na(EcoMonth.df$PrevHeat)
# Sum the number of times on (TRUE=1, FALSE=0)
HeatTimes <- sum(EcoMonth.df$HeatON)
CoolTimes <-sum(EcoMonth.df$CoolON)
FanTimes <-sum(EcoMonth.df$FanON)
I followed up with support on the oddities I was seeing. As usual they followed up quickly and here is what the investigation discovered. + @RyanAtSense for awareness.
Thank you for the additional information on your AC usage. I am seeing that Sense removed a few devices a few weeks ago. We hold Sense to a high standard of accuracy and when Sense thinks that devices are not accurate enough it can temporarily remove them in order to rediscover them more accurately. If these devices were found before, they should be found again given some time. I would encourage you to keep an eye on the AC components that are still active to see if this appears to be both ACs or if it is only one of them.
I appreciate the desire for a high degree of accuracy. I do however question the design choice to outright remove existing devices from view if there’s a question of accuracy. It feels like this choice may lead to a distrust in the product if items are coming and going without explanation of why. If anyone is using IFTTT to trigger an automation based on a device that was temporarily pulled without notice they’re going to kind of be broken until further notice.
I’d much rather see the app both notify the end user as well as move “questionable” devices into some kind of paused state or a temporary device holding area we see in the device list. It could go on explain Sense is doing a little more work to re-validate them and that they won’t be showing in the Now area or updating their data until this happens. Perhaps Sense could even prompt the end user to confirm if one of these devices is on or off when Sense thinks it might be on to help dial in the accuracy.
Thanks for the insight, Brian. I believe this also came up at the local meeting? In any case, it’s definitely food for thought. It is a tough balance for sure, but I like the “questionable device” box idea. It’s always a challenge to get this sort of thing applicable to a broad user base though, as I’m sure you can imagine.
Always feel free to get on the soap box! I really, really do relay this feedback to the teams. It’s super helpful.
Two weeks and no change, though I can’t say I really expected one too quickly given the circumstances. My “missing” air handler and “missing” condenser remain missing or off in some hidden list, and Sense keeps thinking the wrong air handler is on when it’s really the missing one. I’m kind of bummed. Other than our kitchen oven/range the AC systems are probably the largest energy users and I’m sort of blind to them in this state other than the main power meter patterns. I really want to get a good view of their behaviors for my primary use of Sense which was to do generator sizing. Fingers crossed the devices come back during this summer.
It is now two months later and there has been no change for the better with the two AC systems. The devices Sense removed without my consent are still missing and I have no idea if they’ll ever be back. Everything seemed to be working fine until they up and disappeared. If I could just go back to the way things were before they disappeared I’d be happy.
It kind of pains me to say this, but I’m really bummed one of the primary reasons I purchased Sense (monitoring high usage devices for generator plans) has now essentially failed during the time of the year I needed it most.
I pinged support to see if there’s anything they can do on their end such as reset/remove those devices from my account and try to start over.
Can understand why you are bummed. I’ve been trying to study how Sense HVAC detections get out of whack after initially tracking well by looking at what my smart thermostat (Ecobee) is asking for vs. what Sense detects in terms of on-events… Before I share my results, I wanted to point out a couple things that make this a less exact science than it should be:
It’s hard to use Ecobee downloads to figure out precisely what the thermostat is asking for a couple of reasons:
The Ecobee doesn’t say exactly what it is doing. It tells a mixture of what it is doing and what it did. It will identify heat and cool cycles in 5min increments, but doesn’t tell you when it is requesting the fan, or why.
My Ecobee is set to run the fan for at least 5 min per hour (~ 720 per/month), if heating and cooling aren’t moving the air enough. But because the Ecobee ramps up/down the fan with a time offset from the heating/cooling commands, it is difficult to connect the fan being on with a specific heating/cooling command. So the Ecobee count of 5min fan on-events is really a mix of the 5min minimum/hr. cycles plus short heating/cooling (virtually all heating) cycles.
I also have a fair number of dropouts where the Ecobee failed to report in to the mother ship for one or more 5min reporting increments. One has to be smart and make some assumptions about dealing with unknown/NA data to get best estimates from the Ecobee data (I have improved NA handling from previous results)
Sense detections don’t exactly match 1:1 with what the Ecobee is trying to do. Since the Sense can only “see” downstairs/upstairs furnace fans and the AC compressors, it can’t separate out a heating fan cycle vs. a 5min minimum/hr. cycle.
Here’s what I’ve seen since Aug 17 when I reset my Sense data and the HVAC identification looked pretty good. Compares on-events the Ecobee asked for vs. what Sense saw.
Things started out well in Sept 17.
Upstairs and downstairs furnace fan detection went wild, favoring the downstairs fan. Downstairs may be recovering but upstairs is pretty lost.
AC detection has been pretty good until May/Jun, but I’m starting see many more on-detections than Ecobee is calling for, plus a new Mystery AC 3 device has shown up.
I think the big issue here is Concept Drift, where refinements in learning, plus perhaps slight changes over time in the behavior of devices cause them to drift out of the recognition zone.
One additional bit of education out of this analysis - I’ve learned about the high-limit switch on my forced-air, gas-fired Bryant furnace. In pulling the numbers for my tables, I discovered that my downstairs Ecobee was dropping offline far more frequently in the wintertime and seem to drop off often right in the middle of a heating cycle.
Customer support at Ecobee suggested that the furnace high-limit switch is the likely culprit, cutting power to the thermostat when it over temps (160 degrees for my model). He suggested I get a technician to check for impeded airflow or a slowed motor/bad cap. I haven’t measured / verified this and I can’t find documentation that explains exactly what happens when the high limit switch goes on my model. I can find some explanations on Reddit, etc. that say that tripping the limit switch in a furnace drops power to the low voltage transformer to shut off the gas valve, taking out the thermostat in the process…
Anybody else familiar with the behavior of high limit switches ?
I contacted support and they stated the devices while hidden had started re-training yet didn’t redeploy to my account for some reason. They were able to push one device back, my 1st floor air handler, but my other condenser unit is still hidden and its usage is showing up under the “Other” category at the moment.
I’ll keep an eye on things and see if the recently returned air handler device is properly aligning with the actual unit.
OK - All this diddling to compare Sense and Ecobee turn-on cycles had one positive outcome. I was able to determine that my downstairs furnace has been short-cycling for the past year due to a “Limit or Flame Roll-out Switch is Open” error. If I lived in a less mild climate than NorCal, I likely would have noticed the chill, and fixed it sooner out of necessity. I confirmed by running the stage1 heat from my Ecobee in test mode, and the transformer was de-energized at about the 4min point. Looking back at the Ecobee downloads since last August, my median heating cycle runtime was just shy of 5min (300 sec - chart below). Now I need to get the darn furnace checked out.
If you notice your smart thermostat zoning in and out, or going through backup batteries more quickly than you expect, try watching the heating cycles and limit switch more closely.
If I could just go back to the way things were before they disappeared I’d be happy.
