EEVblog #409 - EDMI Smart Meter Teardown

[Frangible]

Even using a 12-bit ADC inside the MSP430, it's still multiplexed, so all the readings are taken sequentially (e.g. Phase A current followed by Phase A voltage, followed by Phase B current, etc.).  Since the readings are not taken at the same time, it kind of messes up the four-quadrant calculations.  If you look at the ADI metering chips such as the ADE78XX series, they are designed with separate ADCs, one per channel, so that all the data is synchronized and can be processed with better accuracy.  I wonder how EDMI gets around that?

[somlioy]

They're starting with those smart power meters aswell in norway soonish. The power company can even remotely shutdown your power if you dont pay your bills!
I've never had the chance to look inside one of those meters.
 
As an electrician I had to smile when you said those were some beefy busbars.  O0
What you see on pictures is the switch for an emergency generator on 350kVA if I recall.
Cables are 400mm², 2 cables in parallell per phase. Those single cables are internal wiring only. Between that switch and the distribution board. Not very flexible cables...
From the power company there's 4, 4x240mm² cables in parallell.
Main breaker 1250A.

[JoannaK]

Even using a 12-bit ADC inside the MSP430, it's still multiplexed, so all the readings are taken sequentially (e.g. Phase A current followed by Phase A voltage, followed by Phase B current, etc.).  Since the readings are not taken at the same time, it kind of messes up the four-quadrant calculations.  If you look at the ADI metering chips such as the ADE78XX series, they are designed with separate ADCs, one per channel, so that all the data is synchronized and can be processed with better accuracy.  I wonder how EDMI gets around that?

Most likely the timing difference is so small they can easily compensate it at software. I can't remember off-hand how fast the MSP AD converter is, since it's been over 4 years since I've done anything with MSP430, but I would say that the timing is not critical issue here.

[komet]

Dave seemed surprised that there were no shunt resistors, but have electricity meters ever used shunts? The old method, as I'm sure everyone recalls, was to drive a spinning disc; that didn't involve shunts either.

I think they probably are actually reasonably accurate. They cannot overread by law (else the power company would be defrauding you), and if it underreads by 1% at 100 amps that could be a couple of hundred $ per year in lost revenue, so there's a clear incentive to measure as precisely as possible.

[kyndal]

looks to me like those "do not connect" terminals are connected to the inputs "before" the current transformer
maby because the power company can't "charge" the customer to for the power the unit needs?

so what i see is.....3 phase of "free power" with no "tampering"  ?  ;o)

be it pretty low Amperage through those little metal clips
hard to tell from the video though.. with it being tossed and flipped..

suppose you could always just tap into the big fat mains inputs...
meh..  just a thought

 *Stick it to the man... or die trying*

/Kyndal

[Bored@Work]

That meter should better be mounted in protected places only. Otherwise the  wireless modules get stolen in no time.

[Saneoc]

    I think that is much more interesting to understand how it can detect bypass of current.
May be could be the case if there would be another current sensor on the N wire, but looks like there is no one. 
 May be it is relying on the fact that there is permanent consumption in the house, so zero readings are suspicious? (this small consumption can be less than one LSB of an ADC)
 

[SeanB]

Most do have a current shunt, normally a big wire loop around the meter case, with a current coil connected in parallel with it. 60A or more ratings requires a thick wire.

As to the generator changeover switch, I did once rewire one in the rain in the mud, after it was dug out of the wall that had collapsed from rain. Needed my ISP back in order to get service.......... Wiped most of the muck off, wiped the busbars with a cloth and quickly bolted it to the fence and reconnected. manual only, as the auto control board had suffered a little from a watery grave, the switch was fine enough. Had to tape a plastic bag over the front to keep it dry, as the seals on the box were a little bad. It was replaced a week later with a new one.

[HLA-27b]

As an electrician I had to smile when you said those were some beefy busbars.  O0
What you see on pictures is the switch for an emergency generator on 350kVA if I recall.
Cables are 400mm², 2 cables in parallell per phase. Those single cables are internal wiring only. Between that switch and the distribution board. Not very flexible cables...
From the power company there's 4, 4x240mm² cables in parallell.
Main breaker 1250A.

Interesting way to mount the lugs to the cables. They don't do hydrolic crimping any more?

[SeanB]

Rivets probably a way to avoid having to use the hydraulic crimper. I still prefer to crimp them, and if particularly paranoid solder them after crimping.

[HLA-27b]

Actually this is an old (think Vikings) boat building technique. But then again the OP is Norwegian isnt't he 

Those interested should check about   Clinker (boat building)   in vikipedia.

also
http://www.flickr.com/photos/antphoto/4288186665/#in/photostream/

[Frangible]

looks to me like those "do not connect" terminals are connected to the inputs "before" the current transformer
maby because the power company can't "charge" the customer to for the power the unit needs?
<snip>

I think those are there for versions of the meter set to work with external primary Current Transformers.  Those types must have their secondaries shorted (through the CTs inside the meter) in order to be accurate - they don't perform very well with burden resistors.  In that case, I suppose the three small connections would be used for voltage sensing on the three phases, with neutral connected.

[SeanB]

The 3 no connects are for final calibration and test. The one screw inside ( the long ones) are removed, so that the current transformers are disconnected from the voltage sensing loop. Then the small terminals are used to apply the phase voltage and the meter power, while the high currents are supplied by a separate transformer that is only low voltage. That way you can test at the full 100A per phase, but do not need a 66kW load to do so, just a small transformer with a single turn secondary adjusted to give 100A through the shunt.

Dave was wondering as to the amount of input protection, but you have to consider that this meter is part of the distribution system, and has to survive lost phases, lost neutral and massive spikes along with massive long duration overvoltages. It has to work both during these and afterwards, and do so for at least 30 years without service. Thus the MOV's and the series resistors for them, along with the 3 phase bridge rectifier before a common mode LC filter feeding the power supply, designed to at least survive one mains capacitor going open circuit without degradation.

Separate power supply for the GSM module is to prevent people tampering by shorting the external power to the module thus turning the meter off. If that output is shorted the other rail will still provide power to the internals.

GPIO and relay outputs are protected as well, relay with a 275V MOV and the open collector optos with 60V MOV's.

On the current transformer side the compensation is done with a RC filter, not in the software. The RC also provides noise reduction. The ZD grey unit will be a small varistor to provide an AC clamp for overcurrent protection. The components there are exactly as seen in the datasheet with the added overvoltage device.

[Rerouter]

Never thought i would see those stand alone modems anywhere else in aus,

they are an RS-232 GSM modem, and have a small issue of locking up around every half year (usually requiring a physical disconnect/reconnect) or at least have on nearly every one of them i have ever met, 

[tom66]

Put a bus-bar the same width and length across the input terminals. This would divide the measured current by approximately half. Any tampering preventing this?

[SeanB]

Nothing detects that. However the metro will have a historical record showing use in the neighbourhood and it flags any meters that are higher or lower than the average for the house and the surrounds for investigation.

When I was living in a block of flats the bulk meter rolled around and stuck for about a year. I freed it by opening the fire connection ( 2 inch pipe) to max and gently tapping the meter case with a 4lb hammer until it started running again. Shook all the dirt out of the displacer and got it running again. preferred to keep the meter than have the cost of having a plumber out to fix our side of the piping if the metro changed and put a new meter. Later the pipe feeding it rusted through, they had to dig back to the main valve in the street to replace pipe, and that valve was leaking and refused to seat ( not surprising as it is a century old gate valve) properly. Fun watching people doing plumbing sitting on assorted cables, 11kV, 400V, a whole lot of lead phone cables and 5 fibreoptic ducts.

[lewis]

Even using a 12-bit ADC inside the MSP430, it's still multiplexed, so all the readings are taken sequentially (e.g. Phase A current followed by Phase A voltage, followed by Phase B current, etc.).  Since the readings are not taken at the same time, it kind of messes up the four-quadrant calculations.  If you look at the ADI metering chips such as the ADE78XX series, they are designed with separate ADCs, one per channel, so that all the data is synchronized and can be processed with better accuracy.  I wonder how EDMI gets around that?

The phase error is absolutely critical for accurate power measurement, a small time difference will affect the result considerably, especially with non-unity load power factors. There's some more info here: http://www.ti.com/lit/an/slaa122/slaa122.pdf

I have another single phase household electricity meter that has an internal resistive shunt with no CT.

I suspect to get the required accuracy and resolution, there are some discrete PGAs built around those LM324s. That's common with energy metering ICs such as Microchip's MCP3905/3906. The meter needs to resolve down to 100mW (<1mA) or less at huge accuracy, and the internal 12-bit converter would be insufficient without input conditioning. Also, it's easy to forget the meter needs to resolve the 20th or even 50th harmonic of voltage and current to get accurate readings, so the demands placed on the ADC/PGA become a bit more significant than at plain old 50Hz.

It's not too much of a stretch to imagine power companies billing extra for poor power factors with all this technology, after all poor PF causes increased losses in the distribution system. (Especially if the whole world is forced to switch to CFLs!). It's easy to imagine being penalised for a new kind of pollution - pollution of the mains. It's started: http://www.lastwordonnothing.com/2012/07/06/dirty-dirty-electricity/

Dave - if you push/hold some of the front panel buttons the display does stuff under its own power.

[komet]

It's not too much of a stretch to imagine power companies billing extra for poor power factors

That's not a new thing, larger customers have always paid extra for poor power factors. The lifts in my building connect to two 1960s-era spinning disc meters - one of them measures reactive power.

[G7PSK]

Bad power factor charging has always taken place for large industrial users, I have installed generating plant for some in the past so that they can shed some of the load from the mains during peak charge times one such company was electroplating and to expand was uneconomical from the mains only due to the high cost of the power factor charge during what amounted to most of the working day so half of the plant was run from two 1200KVA diesel gen sets and all the compressed air came from a diesel driven screw compressor which ran 24/7.

[adh]

Few years ago I had designed something that is actually glorified smart power meter for datacenter applications (combined with remote power switching and environmental monitoring), we had some different design targets than normal power meter so we had ignored issues like phase error but still few points:

I think that the separate isolated power supply is more important for the optional RS-232/485 port mentioned in the datasheet than GSM modem itself. Shorting modem's power rail to turn off meter is probably not interesting attack vector as it's located inside the bottom cover that is also sealed (as it contains actual power connections and is perfect place to bypass the meter). On the related note locating the pulse output terminals and such things under it is probably not smartest design decision as these have to be user accessible in some places. Our device had the whole low voltage part isolated from mains as it was cheaper to source voltage sense transformers than isolating all the I/Os (2xRS232, 2xRS485, 1wire bus and two TTL-level extension ports).

As for multiplexing ADC we also worried about that but found out that it's very simple to correct for that by sampling different ADC inputs in optimal order and interpolating samples, accuracy of the whole thing is still significantly better than what was in our requirements (we really mostly cared about apparent power as that is what you mostly get charged for in DC environments) and probably almost sufficient for grid power meter.

And as for the shunt resistors, I'm actually considering doing new version of this whole device with shunt resistors instead of current transformers because the whole thing will then be considerably simpler mechanically (the thing is limited to about 10A/output phase so there aren't going to be any massive losses). On the other hand that would imply larger mains connected part that has to meet all the mains-related standards (and the I/O isolation problem ).

And for the closing: the whole device is currently in state of development hell as we currently do not have pressing internal need for shifting it from engineering prototype stage to product stage and the whole thing is not as cheap as it originally seemed to be.

[gxti]

I'm also slowly developing a power meter with a large number of current inputs and have come to the same conclusion that it's easier to just use a voltage transformer. The first revision had no isolation with the goal that the data lines would be isolated, but then I wanted Ethernet. So the second revision had the kitchen sink -- Ethernet-enabled microcontroller, isolation between it and the ADCs, and a builtin mains power supply. But the supply didn't work (ugh) and isolators are expensive. So for revision 3 I think I will have only 1 type of input and use a transformer to feed the voltage input(s).

I'm curious about techniques to improve resolution. I can get fractions of a watt just using MCP3903 with a one-pole RC filter/attenuator and no gain, but calculating reactive power is killer because all the noise is multiplied and squared even more than for real power.

[adh]

...
The first revision had no isolation with the goal that the data lines would be isolated, but then I wanted Ethernet. So the second revision had the kitchen sink -- Ethernet-enabled microcontroller, isolation between it and the ADCs, and a builtin mains power supply.
...

I didn't list ethernet in my post because it is interface that has to be isolated no matter what with specification calling for isolation voltage that is higher that what is commonly required between low-voltage side and mains in most electronics. Running ethernet controller directly connected to mains still seems like bad idea (noise etc.) but isolation is not an issue. Important interface that caused me to go with the full isolation route was RS232 and the fact that the fact that it is then possible to meassure all the outputs from CTs and voltage sense transformers as essentially differential signals instead of being Ground/N/PE referenced and thus requiring bipolar supply for ADC or level shifting.

Our first take on this was to have isolated I/O (RS232 only) and using N +/- 2.5V for power which seemed like incredibly bad idea almost instantly.

Main reason for not going with the more obvious way of symmetric supply and some active circuitry to shift output levels into ADC range was BOM reduction, with this design there is not a single analog IC (unless you count line trxs as analog ICs) in the whole thing. Just MCU, few '595s and simple passive networks for signal conditioning (voltage clamp and LPF which were found to be required for proper channel isolation) with everything that can be done in software done in software.

[cengland0]

Even using a 12-bit ADC inside the MSP430, it's still multiplexed, so all the readings are taken sequentially (e.g. Phase A current followed by Phase A voltage, followed by Phase B current, etc.).  Since the readings are not taken at the same time, it kind of messes up the four-quadrant calculations.  If you look at the ADI metering chips such as the ADE78XX series, they are designed with separate ADCs, one per channel, so that all the data is synchronized and can be processed with better accuracy.  I wonder how EDMI gets around that?

I would imagine this doesn't really matter.  With all digital meters, wouldn't there be a sampling rate instead of it being accumulative?  In other words, it must check every millisecond or so and then determine what the usage is at that time and then add that amount of energy usage to the sum.  It cannot do it continuously and be digital too. 

So, if it is multiplexed, it is okay.  It could take a measurement now and determine you're using 7 Amps and that was since the last measurement 1 millisecond ago so you can assume you used 7 * 10^(-3) Amps in that millisecond for that phase.  The faster the sampling rate, the more likely you will be able to catch surges such as air conditioners and refrigerators that have a huge startup cost.

[ve7xen]

So, if it is multiplexed, it is okay.  It could take a measurement now and determine you're using 7 Amps and that was since the last measurement 1 millisecond ago so you can assume you used 7 * 10^(-3) Amps in that millisecond for that phase.  The faster the sampling rate, the more likely you will be able to catch surges such as air conditioners and refrigerators that have a huge startup cost.

Ideally you want to sample V and I at the same instant though.

[cengland0]

So, if it is multiplexed, it is okay.  It could take a measurement now and determine you're using 7 Amps and that was since the last measurement 1 millisecond ago so you can assume you used 7 * 10^(-3) Amps in that millisecond for that phase.  The faster the sampling rate, the more likely you will be able to catch surges such as air conditioners and refrigerators that have a huge startup cost.

Ideally you want to sample V and I at the same instant though.

V is constant, right?  It's 220 or 240.  The product converts the amps into volts with those coils so you're really only sampling the Amps (as volts) and the real volts going into the meter is probably programmed into the firmware or an option set by the tech when it's installed.