the recent LTT video where he runs down all the problems with his house feels so vindicating towards my decision to never do smart control of anything in my house.
@gsuberland did he go with something relatively "proper"/planned or a hodge-podge of gadgets?
I did try a bit of it years ago when it was just getting popular but I quickly realised how many constant little frustrations and frictions it brings. these days it's so much worse because every vendor wants to tie you into their ecosystem, so you end up having to enter into adversarial relationships with products if you want to gain local control over them.
@gsuberland that is why I went with ikea when I bought some smart lightbulbs ten years ago. Their stuff is designed so that you can use it standalone, without a gateway, app or computer.
@gsuberland If i were ever to do "smart" anything it would be proper SCADA based with DIN rail mount controllers and relay modules not consumer junk.
And designed with careful attention to failure modes in case anything misbehaved.
Oh absolutely! Minimal smart, lots of manual is the way to go 🙂👍
@gsuberland I like my smart home, but I'm very deliberate about what goes into it, and the vast majority of cloud-based consumer ecosystems get removed from consideration almost immediately.
@azonenberg I think the main gap there is light switches, where you really do need an integrated solution. but otherwise yeah, actual industrial stuff is the way to go.
@gsuberland My concept there was to have the switch just be a 12/24V GPIO into the controller.
@gsuberland @azonenberg my pet peeve with any sort of “smart” light switch over here in Europe is that they all have way worse haptic design and are more fiddly to use compared to their non-smart counterparts.
@azonenberg hmm, I've never liked that as a solution. I want the catastrophic failure mode to be "the house functions as a regular house".
@gsuberland Yes. Which is why I said "careful attention to failure modes".
For example, relay outputs for wall receptacles would be normally closed, and there would be a watchdog (or manual killswitch) that would shut down the controller on failure and make them passively turn on.
Key lighting circuits would be 2-gang boxes with the second switch (normally left on) controlling power passively.
@azonenberg that's my point though, with that setup the failure mode is "all your lights are stuck on". that's not an acceptable failure mode in my eyes. the only failure mode I want is that the house functions *precisely* the way it did before I installed anything smart.
imo the way to do that is by having every switch point internally contain an SPDT physical switch wired to an SPDT relay in the same way as you would with a lighting circuit with two switches.
@azonenberg that way if the entire smart system fails, or a relay fails, your light switches still function perfectly normally.
@gsuberland They're not stuck on, they're on until you manually turn them off.
There are unavoidable tradeoffs in every case especially in the "I want to switch a wall receptacle with the switch" use case.
@gsuberland @azonenberg
The Shelly Pro's might get close; they have ethernet (and wifi and bluetooth) - but they also have a pair of locally programmable interfaces. So I think you could wire a physical light switch onto the 'SW1' and script it to use that to operate locally;
https://www.shelly.com/products/shelly-pro-1?_pos=10&_fid=34a228ceb&_ss=c
@gsuberland
@azonenberg
There's also a lowtech alternative :~)
@azonenberg but how do you turn them off if the controller is dead and the physical switches are only carrying signal voltages? go unwire it at the board?
@gsuberland When you deploy the system, you didn't know that the switch controlled that receptacle.
The receptacle has to be controlled by a relay somewhere, and that relay has to be independent of the switch (unless you want to lock yourself into a fixed "this switch - that relay" scenario that eliminates most of the benefits of smarts).
The only sane choice is for the relay to be normally closed so the receptacle doesn't go permanently-off on failure. Which means the failure mode *has* to be always-on.
@azonenberg what I'm saying is you have the switch points carry mains just like you would in a regular house, except the switch point contains a relay internally that lets you toggle the lighting state with a smart controller (SPDT switch wired to an SPDT latching relay in series within the switch point itself). if the controller fails, the light switch continues to work 100% as normal. no lighting state changes occur.
@gsuberland This prevents you from remapping the switch to control something else instead of / in addition to the hardwired load (like an overhead light and a table lamp).
That lamp has to fail on or off if the electronics go haywire.
@azonenberg oh I have no interest in that kind of stuff, I just want a simple switch
@azonenberg I will happily give up more advanced control features to keep the basics simple and hazard/hassle free
@gsuberland For plug loads: you use the switch built into the appliance
For overhead lighting loads: 3-position DPDT switch.
Pole 1: common = GPIO, top=24VDC (or other "logic on" voltage, bottom = floating/ground
Pole 2: common = light, top=relay output bottom=relay output
Up position: light wired to relay output, GPIO high. Normal state = GPIO controls relay through controller (but can also drive other relay loads or be switched on/off by other switches, HTTP requests, whatever)
Middle position: light forced off, regardless of relay state (optional: parallel the switch with a normally-open relay)
Bottom position: light forced on regardless of relay state
@azonenberg nah, I hate that. makes it hard to just reach for a light switch without thinking about it.
@gsuberland The goal here is to provide emergency operation of critical loads (so you're not stuck in the dark / unable to sleep because all the lights are on) if the digital control system fails.
This should not be the normal/expected operating mode.
monitoring on the other hand is pretty neat.
I still want to finish designing my whole-house mains power monitoring system. there are some existing projects that do this but I looked at the schematics and they do a bunch of things wrong, especially around the CT sensing.
@gsuberland The worst bit is waking up boiling hot or freezing cold when some part of the system decides not to work.
@gsuberland to some degree I think that’s a choice people make. It is possible to go fully local only, so many people’s idea of a “smart” home is that the light switch is now buried in three different mobile apps rather than on the wall. That’s not an improvement, and if I were in that situation I’d rip it all out.
@gsuberland my headmate built a "smart home" network with openhab and a bunch of zigbee things (only zigbee) and although it took her a bit to set it all up, it's been working essentially perfectly for 2.5 years with no interruptions outside of brief power outages and no ongoing maintenance that we frankly have no effort to spare for
very happy with how it works
@gsuberland My system has been doing very well but it really hinges on choosing the right parts and the right protocols, a lot of unreliable crap out there
@gsuberland she used lightbulbs that treat a power cut as a command to toggle state so the lightswitches work even if the entire network is down, so everything that's safety-critical like lighting has a local override
@gsuberland in retrospect and from seeing others talk about it, i think her decision to go "zigbee only, no proprietary tech, no 'hubs'" was probably what made all this shit work reliably enough that i forget it's even there
@gsuberland (when reading this, keep in mind that for me "standing up to flick a light-switch" can be a multi-hour process, so a smart home system isn't just a toy but something that i need on an everyday basis, and it has to be reliable too)
Edited 1y ago
@whitequark yeah for accessibility it's a gamechanger for sure, if it works well.
but yeah, getting it to work well is... oof
@gsuberland I use devices from egauges.net which can provide revenue-grade metering, and they are also data loggers with a local API, no cloud service involved. I like the data logging aspect as it means I don't lose any data when the monitoring software crashes or gets upgraded.
@gsuberland I desperately want some kind of fancy (probably TDR) mapper that can tell me where the hell my circuits are. But its probably cheaper to just hire someone to do it.
(My mains panel is 70 years old and kind of terrifying, so that needs … work … first.)
@gsuberland Please do share if you get something nice working. I've got some old SEL gear hooked up, but would like a bit higher bandwidth system to see when fun things happen.
@AMS I do actually have a mostly-completed design but it's from way back, when I wasn't as clued up.
the general topology is one mainboard that's fully low voltage, then pluggable fontend modules for doing the mains measurement.
the modules have a pair of ADuM3473s or similar providing two isolated inputs and six isolated outputs in total, plus isolated bipolar power (with LDOs afterward) referenced to mains neutral. then I use direct resistive division and shunt current measurement.
@AMS the isolated inputs are two clocks used to drive the voltage and current ADCs. the isolated outputs pull the serial data back from the ADCs. so each module can do three power feeds.
the idea behind doing direct in-path resistive shunt measurement is that with CTs you get phase error, which is a bit of a blocker for trying to measure power factor or fast current transients.
and the frontends are modular for easy replacement but also because many ADC ICs go out of prod/stock real quick
Edited 1y ago
@AMS it isn't a cheap design (those isolation ICs are like £10 each) but it's still a lot cheaper than buying a commercial one with half as many features.
@gsuberland My thought was resistive voltage and nice closed loop LEMs (or fluxgate LEMs if I find some cheap) for current.
@AMS yeah that's another way to go with it
thing is, I'm gonna be measuring voltage on each line anyway since I want to be able to see if breakers do anything wonky, for example, so since I already have to have some sort of isolated voltage measurement I might as well do the current measurement while I'm at it
@vxo @AMS they end up being a bit more expensive in total, last I checked. with something like AMC3330 + AMC3302 it's about £10 per voltage + current channel (assuming I use the MCU's ADC to measure the amplified output, and I don't need any more opamps for biasing and such), whereas with the ADuM3473 option I can use commodity ADCs and get three voltage + current channels for about £22.
Edited 1y ago
@gsuberland @vxo Yeah, they're like $4 in ones I guess (I'm used to the 100k+/yr pricing at work). 8mm isolated though and nice >80dB noise floor which makes things nice and easy from an isolation standpoint. Being cheap means I'd probably go non-isolated neutral-referenced power and voltag and isolate the current sense. Wifi makes for cheap and cheerful isolation to get the data out.
@kevin decent but pricey.
the two main reasons I want to implement it myself are that I don't want to pay £800 for it and I also want nerdy features like power factor analysis, THD%, frequency logging, and direct waveform capture during out-of-spec cycles.
one of the things I've been considering is ditching CT entirely and doing in-path resistive current sensing so I can get accurate current phase and even store the current waveform.
@HeNeArXn he DIY'd a ton of it, and there's lots of ecosystems involved. he's got WiFi and ZigBee and Z-Wave stuff because he has so many smart switches and such that he can't run them all on one band due to congestion and the meshing being unreliable. but even the stuff within a single ecosystem seems to be full of frictions.
@gsuberland @HeNeArXn Might as well go KNX if your house is that big/complicated