Seeking advice on power distribution in large sculpture

[Involute]

I'm designing a piece of kinetic sculpture that will present 1024 open-frame relays (http://tinyurl.com/yaztd58g) on a vertical panel roughly 10' tall and 8' wide, something like what's shown in the attachment (only about 600 relays in that render, though).  For ease of assembly and transport, the piece will be assembled from 64 modules, each containing 16 relays.  Each module will be controlled by it's own controller (attached to the back), with all the controllers controlled by a master controller sending commands over an RS485 bus.

The relays run off of 12VDC, 170 mA each.  The module controller boards take a trivial amount of 5VDC, converted on board from the 12V.  I've been mulling over ways to supply this massive amount of 12V, 174A, nominally.  At one (unrealistic) end of the option spectrum I could have a single, massive 12V supply supplying the 12V to all the boards.  I don't think I need to list all the reasons this would be a bad idea.  At the other end, I could provide a small (nominally 2.7A) 12V supply on each board, and run AC to all of these.  I'm not wild about this approach either as I don't like the idea of running AC all over the back of the sculpture.  Not that it would be the end of the world, I'd just prefer to deal with low voltage as much as possible for safety reasons.  In between is various numbers (4?  8?  16?) of 12V supplies supplying subsets of modules.  The plan would be to put half the supplies (whatever their number) on one circuit, and the other half on another circuit, so the piece could be powered from two, separate, 15A AC circuits.

Thought I'd check in here for some power distribution advice.  Any constructive comments would be appreciated.  Thanks.

[nctnico]

I'd up the voltage to 48V and PWM the relays so they draw a lot less current when they are engaged. That reduces the peak current consumption (and power dissipation!) drastically. You'll have to tweak the PWM frequency to avoid resonances which are audible though. BTW you do not need to switch the 48V relays when using PWM. Also don't sweat over needing many PWM timers as well. The PWM frequency is likely less than 1kHz so you can use GPIOs from a 10kHz timer interrupt and still have a reasonable PWM resolution.

[sokoloff]

I'd see if I could find DC supplies with "figure 8" or IEC input plugs and run AC on the back of the piece, to be honest. Don't run it as bare wires or Romex, but rather COTS power cords.

Look into local codes and see if power strips (or server rack PDUs) are acceptable. Server rack PDUs are almost surely OK and can give you high density of AC plugs in a nice, neat package.

[Involute]

nctnico:  Good idea.  I'd prefer to see if there's a reasonable brute-force solution first, though.

[T3sl4co1l]

Eh, PWM increases losses and complexity (maybe put an MCU on each module?).

48V is the highest still-friendly voltage, so that's nice.  Still a lot of amps though.

120V sounds better to me; solve the safety issue with mechanical design: insulated connectors, modules, covers, etc.  That's the practical side; on the legal side, high voltage stickers provide CYA.

With 120V input, I'd put a few more per module, so you get a little more economy-of-scale (and efficiency), say using 12V supplies with 5 or 10A capacity.  So, maybe 16 modules of 64 relays each.

Maybe the PCB would be too big then, in which case the PCB could be 16 or 32 relays, and the supply is wired to two or four boards (note: put fuses on the boards, and use current limited/protected drivers, to prevent fault currents starting fires).  Whichever -- it's more of a mechanical decision at that point; the electronics can be whatever.

Note that's still 17.4A at 120V (at 100% efficiency).  That's more than a standard 15A outlet, and you'll need PFC to even get that much (most off-the-shelf supplies have a power factor around 0.6, limiting you to about 1.2kW on a 15A circuit).

This thing is going to get HOT if it's doing continuous duty (all channels on)!

You may even want a 240V circuit.  Depends what the installation site can provide: two 120V 15A circuits, or one 120V 30A or 240V 15A circuit?  (208V 15A would be acceptable as well.)

Also, don't forget to filter things carefully.  Slew-rate-limited drivers would be kind here.  It's not that the relays are on long cables, but the power turning on and off has to be communicated up to the power supplies at some point, and that gives an EMI emission surface of multiple boards.  At the very least, that can screw up your own communications (RS485 is quite robust, but I can see an -- admittedly, /very/ poorly designed -- implementation causing problems!), let alone anything nearby.  If you use ribbon cables for communication, don't forget to interleave signals with grounds (or pair up differential signals like RS485, and surround those with grounds on either side), and connect all grounds to GND plane, at the header.  If using multiconductor cable, connect screen/shield to GND on both sides.  May want to put snap-on ferrite beads on the cables.

Tim

[Involute]

Thanks, sokoloff.  Given the vibration this thing will be subject to, I think I'll need more positive mating than a friction fit cord.  I'd absolutely use some "legit" cord for the AC, but at the PSU end I might replace whatever connector is there with, say, a Molex plug with a snap fitting. As for the AC end, since the piece may occasionally be operated outdoors (left that out, heh heh; it would never be exposed to rain, though), I'm thinking of standard AC duplex outlets with in-use covers (http://tinyurl.com/yc2y9vko) to at least keep the dust out of the sockets.

[Involute]

Thanks, Tim.  I figured I could get by with two separate 15A, 120VAC circuits, even without PFC.  Of course, PFC would help even more.  Regardless, I've always assumed I'll need two circuits.  240VAC won't be an option.

[Siwastaja]

Totally agree with PWM from 48V. Especially if it's already driven from a microcontroller; it should already use a half-bridge (i.e., a freewheeling diode and DC link capacitance), so no extra complexity generated. Transitions matter, they would just happen more frequently.

As the required PWM resolution is not high (for example, a 3 or 4-bit counter will work), software PWM will work if hardware pwm channels run out.

48V would be fairly optimal because it's a safety extra low voltage, and now your currents would be small enough to go through fairly standard connectors.

The PWM should be at over 20kHz to get rid of the audible sound - unless, of course, you want to explore another artistic direction on this and make some music with it. Whether this causes too much iron losses in the relay, should be tried in practice.

Another trick is to reduce the "keep" current. To make that nice "click", you indeed want to supply the relay with full 12V. But it makes little sense to keep it fully powered just to warm it up - it will most likely hold just fine at 3-4V, especially since it's not actually being used (so the force on the contacts for minimum contact resistance is not important). In fact, many bigger/more expensive relays contain an integrated PWM circuit or "economizer" which reduces the drive after the initial pulse. This trick could easily reduce your total power consumption to about 1/4th, which would be a huge win!

This plays along well with the PWM thing. For example, if you run it from 48V, just start it at 25% duty for, say, 50 milliseconds, then go to, say, 5-10% duty. Of course you'd need to test this out how it exactly works best. Of course, if all the relays switch at exactly the same time, the worst case consumption is still there. But since it's only short term, cabling and connectors can be rated for long term average. Remember to add fusing based on cable/connector ratings, however.

[tpowell1830]

Okay, so I would use 2 or 3 12 VDC supplies @120VAC input each. Assuming that 60% efficiency  (you should try for at least 75% efficiency) worst case, this would get you to the solution. 17.4 amps/0.6 =  29/2 = 14.5 Amps each 120 VAC input. As far as running AC all over, you don't have to do that because your relays are DC. Run the wires of your 12 VDC supplies to a bus to your control circuits. From there, run wires @12VDC to your relays. Use SO (16 AWG regular power cord) cable to do this. This entire wiring nightmare will be quite heavy, so I hope the sculpture (structure) is very robust.

Use proper care to cover the bus with a plastic or non-conducting enclosure. You may need some cooling fans, you can get some fairly quiet fans with some research. Depending on how fast you are switching relays on and off, will depend on whether they will over heat. Give them plenty of rest time between cycles. Also, the relays may need a reverse diode on each one to prevent high voltage spikes on each cycle. Sometimes relays come with built in diodes.

I also like the idea someone said about using 220 VAC inputs, since this will divide your required input current, although this is probably not an option on a wall socket in a museum or gallery.

Hope this helps...

[nctnico]

The PWM should be at over 20kHz to get rid of the audible sound

That isn't going to work because at 20kHz the iron cores won't be able to keep up and due to the high inductance you won't get any current through the coil. When using PWM for a relay you have to think about 1kHz max and probably much less (a couple of hundred Hz).

[james_s]

Server power supplies are readily available for little more than the cost of shipping. These are typically 500W-1kW with a single 12V main output and a smaller 5V standby output that can usually deliver a couple Amps. Additionally most of these units have current sharing pins so they can be run in parallel or with a simple modification you can run many of them in series for higher voltage. I use a series wired pair of 675W IBM power supplies to power one of my RC airplane chargers. With huge numbers of orphaned very high quality server PSUs on the surplus market there is nothing else that can compete.

[Involute]

Thanks, James.  I'll look into these.

[T3sl4co1l]

The PWM should be at over 20kHz to get rid of the audible sound

That isn't going to work because at 20kHz the iron cores won't be able to keep up and due to the high inductance you won't get any current through the coil. When using PWM for a relay you have to think about 1kHz max and probably much less (a couple of hundred Hz).

There's average DC, it'll work just fine.

What may not work well is the eddy current in the core, which looks like a shorted turn -- a fixed resistance in parallel with the coil inductance.  Pushing ~48VAC through this, will definitely heat it up.  Though it's not clear by how much.

Tim

[nctnico]

The PWM should be at over 20kHz to get rid of the audible sound

That isn't going to work because at 20kHz the iron cores won't be able to keep up and due to the high inductance you won't get any current through the coil. When using PWM for a relay you have to think about 1kHz max and probably much less (a couple of hundred Hz).

There's average DC, it'll work just fine.

Been there, done that and it doesn't work very well to say the least.

[Towger]

Many years ago I saw  something like this.  It used old telephone relays and was controlled by several Apple 2Es.  Just be aware it was very noisy, which may be a problem.

[Involute]

Just be aware it was very noisy, which may be a problem.

That's one of the goals.     They won't all be firing at once, except occasionally, for a climax.

[james_s]

I should also mention there are large 48V server power supplies available, these typically are around 2kW and require 240V input but it's an option if you decide to go with 48V.

[Involute]

I'm not going to have access to 240V, plus I'm leery of using used equipment in this thing.  And, 2kW is about what the whole piece will require, in raw power.  I think it would be better to divide it up among several smaller supplies.  I'm leaning toward eight now, so figure 3kW/8 = 375W each, which should offer plenty of margin.

[james_s]

Server PSUs are extremely reliable, I'd generally trust a used one over a brand new lower cost PSU, new they typically cost a few hundred bucks each. Occasionally you can find brand new ones that are surplus, servers get decommissioned when they become obsolete and the spare power supplies get orphaned.

[TerraHertz]

Sounds like it will be a cool thing to see and hear. Actually, if you can pulse modulate the relays with individually variable frequency in the audio range, it could sound awesome. Different frequencies drifting around in the array...
Ha ha, phased array sound pulse generator art installation, able to kill randomly selected audience members at decent range. A whole new art experience. I hope you're not working for the military.

I expect you're aware of these:

'A million times' by Humans since 1982


Kinetic Art - Dynamic Structure 29117

However you partition the power supplies and relay driver boards, you'd probably better keep the two partitionings the same.
ie for each set of relays from one power supply, one (or more) driver board(s) run from that same supply.
Any communications outside those partitions, via electrically isolated channels.

If you make the power domains as vertical columns in the frame, then the heavy power supplies can all be at the bottom, with minimal cabling. And the frames holding the relays, a power supply and driver board, could all be identical, stand-alone units, that you can bolt any number of together horizontally.

[capt bullshot]

PWM and stuff has been mentioned here. And your setup will draw quite a lot of power if all the relays run off their nominal voltage.

One way to reduce the power demand dramatically is to reduce the relay coil voltage once they're engaged. A typical relay will engage at maybe 70% of the nominal coil voltage (depending on temperature and whatever), so the nominal voltage is used to engage the relay. After a few ms (depending on the particular relay) the coil voltage is reduced to 25% ... 50% of the nominal voltage, this is enough to keep the relay engaged, but consumes way less power.
So if you design your controller board accordingly, you turn on the relay by applying 12V to the coils, then reduce the coil voltage to e.g. 5V. You'd just build a somewhat larger 12V to 5V switching mode converter per module, and your total current consumption is reduced, easing the power distribution problem.
Total power depends on the switching pattern and frequency of your sculpture. Capacitors on each module board could supply for the turn-on peaks, if the switching frequency isn't too high and the relays don't switch on all at the same time.

Of course, some kind of uC generated PWM could also be used to supply the relay coils with 12V for turn-on and 5V for keep-on on a per-relay basis.

[Involute]

> Different frequencies drifting around in the array...

Though the original concept is for a square array, the 1024 relays could just as easily be arranged as a single row 250' long, or anything in between, including a horizontal ring the listener would step into the center of.  These other arrangements would have an even greater spacial component.

> I expect you're aware of these:

Yup.

> ... Any communications outside those partitions, via electrically isolated channels.

That's the idea for the moment, though the RS485 cabling would include a common ground line so all transceivers can reference the same ground.

> If you make the power domains as vertical columns in the frame, then the heavy power supplies can all be at the bottom, with minimal cabling.

That was my original thought, too, though it later occurred to me I might want to put them half way up since there'd be 8' of cabling from the bottom to the top vs. 4' from the middle, reducing any voltage drop.  Might not matter with fat enough cables, and the relays won't be that sensitive to a volt or two anyway, I imagine.  It'll be easy to test when I have some of the pieces.

[Involute]

> You'd just build a somewhat larger 12V to 5V switching mode converter per module

In this context, I use the term "module" to refer to a panel with 16 relays and a controller board, possibly with a power supply (though, increasingly I'm thinking of moving the power supply off the module into fewer, larger supplies scattered around the assembled piece).  Having to supply two supplies, plus switching for each individual relay, would probably exceed my complexity tolerance.     I could achieve the same thing with PWM (as you and others have mentioned), moving the problem into the software domain, but I haven't thought about THAT complexity either.  I'm currently planning on using an AVR ATMega32A which has 28 IO lines.  I know it doesn't have enough h/w PWM resources to handle 16 independent relays, but it might be possible to do it in s/w.  The MOSFETs I'm now currently using can handle up to 60V, so I can design the h/w assuming big, 12V supplies and later switch to smaller, 24V or 48V supplies if it turns out I have the bandwidth for PWM.

[BrianHG]

To save current  on a 48v relay without PWM, once a 48v relay is engaged, it will hold fine with around 15v, a cheap trick would be to use a series 32v zener diode with a 1000uf cap and 3.3k ohm resistor all in parallel on that diode.  When first turned on, the cap has a 0v charge sending the full 48v to the relay.  As the current of the relay fills charge in the cap, it will charge to 32v and stop there by the zener action leaving 16v to hold the relay at a fraction the operating current.  When turned off, the 3.3k resistor will discharge the cap

The 48v version of your relay coil's resistance is 1.2Kohm, or, will consume 40ma when on.  Or around 2 watts per relay.

When powering it at 16v after turn on, it will only consume 14ma, but remember, with my linear step down, this 14ma drawn from your 48v so your power drop will be from 2 watts per relay to 0.7 watts per relay.

With PWM on each relay, you will have a nice radio transmitter in each relay coil if you have no filtering, but, that 0.7 watts will drop in half.

There may be other cheap tricks if you want to use cheap DC-DC converters from China with an adjustable input giving you a filtered DC output which sadly will cost less buying the inductors and caps and assembling them on your own PCB, but if you are already making your own PCB, a MCU solution would be a better fit and if price isn't a concern, go right ahead and DC filter each output from your PWM channels.

1 of these supplies should do the trick if you are careful of how many relays are simultaneously switched:
https://www.jameco.com/z/SE-1000-48-MEAN-WELL-AC-to-DC-Power-Supply-Single-Output-48-Volt-20-8-Amp-998-4-Watt_2101577.html
2 if you want to never run into a potential overload.

Note that all the big supplies have fans, so, they do make noise when hot.

If you don't want a power supply fan, I will have to make a new recommendation, but, it will mean going to 240v, but, no fans.  Though, the relays will still radiate their specified amount of heat.

[ajb]

I'm currently planning on using an AVR ATMega32A which has 28 IO lines.  I know it doesn't have enough h/w PWM resources to handle 16 independent relays, but it might be possible to do it in s/w.

The nice thing is you don't need much resolution at all.  A timer ISR running at a couple of kHz or something should do the job just fine.  If you round your 'hold' duty cycle to the nearest integer fraction 1/n you simply have 1/n of the active relays on at any given time and your PWM frequency winds up being F_timer/n with a current waveform that is practically DC.  You can slim down the ISR quite a bit if you compute a lookup table whenever the state of the module changes, then the ISR just has to throw the next table entry at the IO pins each time it runs. 

[Involute]

Thanks for thinking this through, Brian.  They're 12V relays, but I get the idea.  I could probably overdrive them briefly with 24V, but 48V might significantly reduce their life expectancy, even if applied in brief pulses.  Regardless, I'd still need a supply capable of supplying full power, if only briefly.  So it seems to me that this approach, as well as PWM, would both be solutions for reducing heat (due to lower average power), but not the size of the supplies.  The relays won't be energized for more than 5 seconds at any time (and, generally, more like half a second), so I'm not sure heat is going to be an issue.  I don't think fans will be a big deal, since this thing is going to be making quite a racket on its own.

[BrianHG]

Are you saying you already have the 12v relays?
48vdv, 120vdc, 120vac versions exist.
If you are stuck with 12v versions because you already purchased them, the approach will be different.

If they are only going to be energized for a second or two at a time, with multiple seconds off, please provide us with the average duty cycle.

Maybe all you need is a good capacitor in the high 47000uf for around every 10 relays on each controller board which will have enough kick to switch on the relays, using a 1kw linear 12v unregulated supply.  1 beefy toroid should do the trick with a few stud diodes and 100000uf cap driving a whole bank of relays.  Don't bother with the PWM as this stage.  This type of supply and all the DC filter caps properly wired and laid out will send 2-3kw power surges without even noticing a thing.  No voltage regulation required for the relays, just use a 10v to 11v transformer.  For the MCU driving the transistors or logic level mosfets, just add a series 1N4002 and a big cap + 7805 which should handle any deep power sags.

[nctnico]

Thanks for thinking this through, Brian.  They're 12V relays, but I get the idea.  I could probably overdrive them briefly with 24V, but 48V might significantly reduce their life expectancy, even if applied in brief pulses.

Definitely not. The coil of a relay is an inductor so when power is applied the current will ramp up linear. So if you cut the power at the right time (=PWM) it is as if the relay is powered from a lower voltage. In one of my designs I used a 12V relay from a power input which can range from 20V to 80V. Using PWM and constant power control the relay is perfectly happy.

[BrianHG]

Thanks for thinking this through, Brian.  They're 12V relays, but I get the idea.  I could probably overdrive them briefly with 24V, but 48V might significantly reduce their life expectancy, even if applied in brief pulses.

Definitely not. The coil of a relay is an inductor so when power is applied the current will ramp up linear. So if you cut the power at the right time (=PWM) it is as if the relay is powered from a lower voltage. In one of my designs I used a 12V relay from a power input which can range from 20V to 80V. Using PWM and constant power control the relay is perfectly happy.

This is clearly a misunderstanding.  I was telling the OP to buy the same style relays with 48vdc coils.  The 48v version has 1.2kohm coil which makes for a much lower current allowing for much thinner conductors and a much larger acceptable turn on voltage.

This shrinks drive transistors, their heat, allows for my zener diode and cap trick.

[Involute]

> Are you saying you already have the 12v relays?

No, they're just the ones I initially zeroed in on due to cost, appearance, size, and preferring 12V to other choices.  The 12V preference was due to the realization that I'd be generating my small 5V logic supply on each PCB from the relay supply.  The current would be small, so a linear regulator dropping 12 - 5 = 7V wouldn't get too hot, and would be cheap and easy.  Increasing the relay supply invalidates that and pushes me into a small SMPS.  Not the end of the world, so if going to a higher voltage relay supply is warranted, so be it.  As I've said previously, I don't think heat will be a major concern, due to the low duty cycle of these relays, so I'm still not convinced I need to go to 24V or 48V, unless I'm missing some other benefits.

[KrudyZ]

Where will this sculpture be living?
You might have some serious issues with EMI, jamming radios, etc. all over the neighborhood...

[Involute]

> Where will this sculpture be living?

Unclear at this point.  I'll be building it on spec through crowdfunding.  Before the campaign, I plan to contact music festivals (Coachella, Lollapalooza, Electric Daisy Carnival, etc.) to see if they'd let me display it.  Ultimately, there's a new, interactive space opening here in Los Angeles later this year that expressed interest in a smaller version, though even that was beyond their budget.  I'm pretty sure they'd be willing to take this one for a deep discount (or even for free), which i'd be able to offer, since it will have already been paid for.  I realize crowdfunding's a long shot for something like this, but I need to go through this exercise just to figure out how much it will cost, and if it will work.  I plan on building one module for testing.