Ongoing device detection issues

Starting to get OT but the base calc for AO shouldn’t be doing that even for a 1w delta for a 2w actually-always-on device. Are you sure the 5w minimum is constant?

Recently I’ve seen that firmware updates threw off the AO listing because of the nightly reset treating the devices as “going to 0w”. Mine (on smartplugs) have all returned to “normal” now after 24hrs.

My smartplugs are also pretty rock solid, even for 1 and 2W consumers.

@samwooly1 @ixu @kevin1

I talked with the platform team and the interaction between AO and smart plugs is as follows. The threshold for a device to be moved from AO is 15W, except in cases where the wattage change is sustained for 10 seconds.

So, in your case @samwooly1, it makes sense if that 6W is held for >= 10 seconds. Otherwise, that should not be happening and you should reach out to Support.

1 Like

Very tired of the lane excuses why Dense Sense can’t identify major loads in my house.
I have two of the same model AC condensers and Sense calls one “Other”.
Just get the high energy loads correct and I would be happy.

Have two identical models really means nothing. They can be enough different to not be recognized.
They could have a different compressor in them, this happens a lot when the same model can use several compressors.
I had a detected water heater that was 2 weeks old and had to be replaced due to a factory defect.
I replaced with the identical water heater but had to wait for a new detection. There was enough difference to make the original detection worthless.

1 Like

I’ve just been reading about otter.ai’s advances in speech recognition. Are we in the second or third decade of usable recognition? I haven’t found that the Sense folks offer excuses … they know it’s a very hard problem, especially for a relatively small group. Consider the relative funding for speech recognition!

I hear you … but I wonder how many more decades before any AI will be able to distinguish, say, 20 voices on a bad conference phone. Some of them perhaps from identical twins.

Noisy conference all is a great analogy, but skips the fact that Sense doesn’t look at “conversations” or even individual voices talking continuously. It only looks at start and stop transitions. So, for example, my always running computer center is in the “never detect” (continuously on) category.

It’s a very very difficult problem and that it works at all is pretty amazing. The biggest problem is that expectations set by their marketing are way beyond what their technologists can realistically deliver.

True, though I would push back a bit and say that it’s all a question of scale.

For any given start/stop there is a necessary period of pattern matching that is not exactly “instantaneous” … and no-doubt that period will be extended as the processing and ML matures. So within that period looking for distinguishing characteristics is the same problem on a shorter timeframe.

Like on the conference call a few people yell out “No!”.

“Yeah, yeah, I heard but who all said that??”

Sense is definitely looking beyond the short transitions because they are detecting some EVs that have slow ramps as a single detection. It’s just that those detections have to be done outside the special, highly-tuned capabilities of the Sense monitor. I think long on and off transitions have to be sorted out back at the mothership using the Sense half second data.

For example, Sense detects my Model S

Once upon a time (June) it detected my Model 3, but gave up when Tesla updated the charging cycle via new firmware.

image

1 Like

From all the responses I’ve read from Sense, they were kind of surprised to see that (and electric cars are becoming a major market), so they have special cased such signatures. That’s one major reason that each car and even each charger requires a package specifically for it. Painful when trying to write generic software.

Since dozens (hundreds/thousands) of modern devices (like most modern refrigerators for example) also don’t do bang/bang on/off, that changes the whole ball game.

1 Like

I would credit the Sense techs with understanding that an electrical device load can be bang/no bang/smooth/rough/ramped/all-over-the-place and that ultimately the on/off aspects to signatures is only a small part of what Sense needs to do and does do.

Part of the challenge is that until there is enough data the larger scale patterns are a bigger challenge.

Same compressors. Same fan motors.
Same, same. One is detected the other is “Other”

And some identical twins are truly identical in every way, and others are not. The same motor and same fan may have the same part number, or spec, but may not actually behave the same the way that Sense sees them.
And on the flip side, if they were truly identical, how would sense know which was which? There are many (myself included) that have both air handlers identified as the same. Which ever turns on first gets tagged. Then if the second turns on while the first is running, it goes to other. Check out this blog post about AC units in particular. I know it doesn’t solve you problem, but I think it sheds some insight.

They may have the exact same numbers but will have different signatures when turning off/on.
You would be surprised how slightly different internals that would be considered normal anomalies will affect detection.
That’s why I was telling you about my identical water heaters. I have even lost detections because I changed a breaker in the panel. I changed it with the same amperage but different brand and lost everything detected on that circuit.

The windings in a motor like a compressor could have a few more wraps of copper in one and that would make them look different

I’ve had mine for about 9months, and I’ve given up.

It’s interesting, and has a ton of potential, but the inability to help tell Sense when things go on/off is frustrating.

I get there is a lot of aspects to detecting, especially when one is not monitoring each specific outlet, but there seems like there are things we could be doing to better help out their logic.

Each house would be different, since just the wiring in a 1940s home is characteristically different than one built in 2018…I’m guessing the exact same fridge in both houses look different, but I could be wrong. Heck, depending upon how the house is wired, I’m guessing the exact same fridge in two different houses built in 2018 is also different…probably close, but (again) I could be wrong.

My suggestion would be to:

  1. do a circuit panel learning
  • turn off every circuit breaker, and then bring them online Individually, having me indicate which breaker # it is, the size of the fuse, and every known item on the breaker
  • if I really wanted to learn the signature of just the wiring, I’d have the home owner unplug everything, and use a supplied Sense Learning item that when plugged into each 120/240 outlet helps sense know that outlets pattern (I’m assuming that with a Sense known test signal, this is possible)…track by outlet # upper/lower
  • I’d then have the home owner indicate what is plugged into that outlet (make & model), if that outlet is switched, and/or if nothing is plugged into it
  • empty outlets I’d ask for common uses (typically where vacuum goes? Or the waffle maker?)
  • then I’d have Sense guide me bringing the whole house online, which breaker in what order, including having me wait (30min if needed) before the next breaker is flipped on
  • this seems like would be a great baseline for Sense to start from…just to know the signature of the house wiring…MY unique house pattern
  • use that learning device again on all outlets when the whole house is online (back to outlet # upper/lower) to get an after pattern of my house
  1. device Learning Pattern wizard
  • based upon the type of device, I’d have the homeowner perform various tests
  • fridge ex: open and close doors (light signature), run ice maker & water dispenser
  • micro ex: stove light (high/low), stove fan (high/low), boil water for 2min (at various power levels), etc.
  • same concepts for ceiling fans, garage door openers, dishwashers, dimmer lights, etc
  • when I plug my RV or boat into my garage to charge their batteries
  • small appliance (waffle iron) tests
  • technology charging (laptop, iPhone, etc)

All of this wouldn’t be the final signature when the whole house is alive, but I’d assume would help MY Sense learn MY house. I’m willing to spend the time training since it benefits me knowing what everything is doing.

At the moment, MY Sense needs help. It has recognized my dryer as 2 unique devices, an AC when it first starts (the spike), and then as a dryer when it runs its cycle.

It has yet to find my furnace, but did find my AC. In fact, it thinks I have 5 AC units (only have one)…after sleuthing, dryer was one, portable dehumidifier was another, stand-up freezer was another, and I can’t locate the 5th yet (which is where my curiosity on could Sense let me know which circuit it’s on if it went through a circuit learning test). 5 AC units? Really? Makes me wonder how smart the Sense logic truly is when it doesn’t recognize (and inform me) that more than one AC has been found…I don’t know any homes with more than one AC. Maybe a Zip code of multimillion dollar homes, but c’mon, put some common sense smarts in the app too!

For me, I’ve not really gotten the monetary value I’ve invested in this. It now just sits and listens, I rarely open the app or see where I’m at. Why bother when I really can’t tell what (beyond which of my 5-“AC” units) is sucking my power?

Mike

1 Like

I have given up as well - it’s an expensive meter reader. It would be so easy for the engineers to implement MSBENDTS assistance idea but my guess is the engineers/management think their little machine learning algo is what will get them promoted or some big IPO money - i haven’t seen a single ML solution work in the wild without both good human design and realtime human feedback to drive refinements/learning.

They are going to have to find a way to incorporate “product wishlist” items with what their goals are for Sense. I understand they have their hopes and dreams for Sense but they can’t get there without the customers.
When my wife opened her business,
We spent tons of money on marketing and advertising. It was all wasted, word of mouth will make or break a business. We haven’t spent a penny on advertising in 10 years, customer service and satisfaction are the priority.

I’m still happy with my Sense experience mostly, but i do understand where others are coming from.

We have incorporated many Wishlist items from this forum. I’ve posted the link before and shared with your personally. There’s currently 36 in the Released Features subcategory. More could be added, but people are combining requests into a single thread, thus I cannot wholesale move the thread over.

https://community.sense.com/c/product-wishlist/released-features

Everyone:

We absolutely understand the frustrations, but this thread is veering off the rails. I would encourage everyone — new users and old — to read through the Community Guidelines: New Sense User? New Member? START HERE.

This forum was built as a place to for users to help each other and to pass feedback on to the Sense team. Infighting and complaints do not add to the conversation, whether they are directed at other users here or the Sense team.

Please, for the sake of all users and especially for the sake of new users who just come here to get some help, keep things constructive. If you solely want to voice an opinion — positive or negative — to the Sense team and not join a larger conversation, shoot me a PM or send Support an email. That is a better avenue to reach us directly.