Procooling's First monthly Electronics Design Corner
 

The plan
This is the newest of the half-baked ideas here a procooling: a monthly electronics project designed by committee in our forums. I am going to decide upon an electronics project every month, and then we are going to agree upon features and design parameters for this project. Then, we will come up with one or more than one design. I am going to put a cap of ~$75US per month on parts (but I am not counting resistors, capacitors, breadboards, solder, wire, or anything that people probably have around if they are building electronics projects at home) because I am paying for this out of my pocket. So if it is an involved project I may not be able to make as many variants as if it is a simpler project. I have never etched a PCB before, but if it is agreed that this is the way to go then I can either get a kit or get someone in the EE department to have the pcbs made for me. Unless we agree that a design either does not meet the goals or has a major flaw, I will build all of the completed designs (that I can afford to make), compare/contrast them, and give a final analysis. Parts lists with digikey numbers, pcb drawings, and all other information needed to reproduce the project will be posted on Procooling. I think that we should take 2 weeks in the discussion/ design phase and then I will have 2 weeks to assemble and test the designs. Maybe, just maybe, I can track down a sponsor for this article series to get some goodies for contributors (like the Pro/Geeks section). If not I will just sign you up for pr0n via e-mail or something.

The project
Peltiers seem to be gaining in popularity with water coolers every day. Sure they are expensive to run and potentially dangerous, but they get CPUs COLD and are a lot less of an investment than a commercial phase change system if one already has a watercooling setup. The dangers of peltiers aren’t inherent, but they add complexity to a cooling system and can fail catastrophically when care isn’t taken. I can’t count the number of threads I have seen on forums that detailed someone’s spectacular meltdown from a pelt mishap. These are usually due to: pump failure or running the peltier without the pump being on, but I am sure you guys can imagine some other scenarios. Peltiers on graphics cards seem to be the hip thing now, and the thought of burning up a $500 vid card makes me cringe. So I am proposing that we develop a comprehensive control system for peltiers that makes them more suitable for use in the system of an absent-minded person like myself.

The goal here is peace of mind while using a peltier-based system. The features that came to my mind immediately are:

1) Turns peltier PSU, pump, and fans on and off with the PC so that there is no danger of running the pelt without pump.
2) Shutdown if temperatures exceed a certain point (the cold plate seems a logical spot to measure to me).
3) Shutdown of system in the absence of flow

And I can think of one other feature that would be interesting to add: A delay for PC boot until the cold plate is cold (a la phase change setups).

Ok then let's get this party started!

Nice.

We'll have to assume that the Pelt PSU is seperate from the PC's PSU.

As a power source, the PC's PSU has a +5 line that's always on (+5 standby), as long as the switch is on, so we can tap into that to power this circuit.

So the objectives are:
1) Turns peltier PSU, pump, and fans on and off with the PC so that there is no danger of running the pelt without pump.
2) Shutdown if temperatures exceed a certain point (the cold plate seems a logical spot to measure to me).
3) Shutdown of system in the absence of flow
4) A delay for PC boot until the cold plate is cold

#1
We might not have enough power to drive a DC fan off of that standby line, so it'd have to be driven from the Pelt's PSU, or be left off until that coldplate temp (#4) is good. The pump can still turn on though (if it's an AC pump).

#2
We can either use a thermal switch (easy, bulky?), or figure out how to setup temp probes. I prefer the temp probe, it seems more interesting.

#3
Let's try to avoid the OC flow sensor, the one with the cork plug ;) . It shouldn't be a big flow restriction, whatever we come up with.

#4
If we're going to build a temp probe, that's the way we should do #2 as well. Do we want to tap into the "Power_OK" line from the PSU, or just cut off the "Power_ON" line? Could we add a solution, alongside, to allow for a soft shutdown? that would mean setting two temp levels, one for a soft shutdown, one for a hard shutdown, unless we just implement a delay.

Ok, I might as well join into the fray ;)

1) Turns peltier PSU, pump, and fans on and off with the PC so that there is no danger of running the pelt without pump.

If you want to use #4, it seems this needs a bit of modafiation. Turn everything but the computer itself on until temps stablise, then allow the computer itself to turn on.

2) Shutdown if temperatures exceed a certain point (the cold plate seems a logical spot to measure to me).

Excuse the ignorance, but what temps are we aiming for here?

3) Shutdown of system in the absence of flow

How would you like to measure this? Bladerunner did an article on flow protection here. Are you thinking a similar consept?

4) A delay for PC boot until the cold plate is cold

I think that the grey wire on the ATX connector is the one we want to look into. A simple extention for the ATX connector could be hacked into submission ;)

The mobo doesn't turn on without the power_good signal, correct? Yet the powersupply will be on. So it doesn't matter if the pelt/fans are on the seprate supply.

Now, let me ask, what do you want to happen if coolant ceases to flow? Should the circuit just kill the mains (120v) feeding the comptuer, or try a soft shutdown? The two levels suggested by ben make sence.

So, a recap.

1. User presses power button
2. Comptuer PSU turns on, actavating relay to turn on pelt PSU and pump
3. circuit automaticaly powers down (hard power down) if flow is not detected (continues to monitor this).
4. circuit begins monitoring temps
5. circuit waits until temp probe reaches x deg C, then gives mobo pwr_ok.
6. Computer powers on.
7. If temps reach above x deg C, circuit attempts soft shutdown.
8. If temps keep rising, citcuit automaticaly switches off mains, and triggers a led indicator so the user knows there is a problem.
9. On shutdown by the user, the relays to the pelt and pump are turned off, shutting down the cooling system with the computer.

Perhaps leaving the watercooling system for a small ammount of time after the comptuer has been powered down would be another feature to add. A suitably sized cap in line with the pelt/pump relays would work (although this would require the fan being on either the pelt PSU or the 5VSB line...

Edit: fixed pwr_ok color

To delay the boot, the circuit can power on and short out the reset switch on the mobo for the time one wants to delay the boot.

Also consider that there is a LOT of redundancy in the current goals; are they all needed?

If pump fails then temperature would rise. System could be shut down by this rise and therefore no need to monitor for flow.

The REAL goal is peace of mind while running a big beefy peltier based system. How to get there most elegantly and effectively?

For example, KnightElite operates simply using a surge protector that I wired a 12V relay into. The pump and pelt psu are plugged into that and so they turn on and off with the PC. Is this enough? Not especially elegant for sure. The other end of the spectrum is the setup that Asetek (or even Corsair) use as controllers for their systems. I was thinking that a homebrew reproduction of those might be of interest and value.

Don't bother with that, just hack the ATX connector so that the motherboard doesn't get the pwr_ok/pwr_good signal until the circuit is happy ;)

Just like on the vapochills (grey wire=pwr_good/pwr_ok) :D

Edit: oops, just looked over the atx spec and pwr_ok is grey, not purple. Purple is 5vsb. Oops :rolleyes:

Reading through the ATX spec (http://www.formfactors.org/developer...12V_PS_1_1.pdf), section 3.3.1 says that is is just a 5v ttl output that tells the mobo weither the power is good or not :)

I am just wondering if the mobo will indeed refuse to start up without this input :shrug:

Pardon my ignorance, but what does Asetek or Corsair have that's of interest?

Otherwise you're right, the flow is redundant.

Whatever we do, let's stay away from trying to control the Pelt, to maintain a temp, OK? It's just too complicated to try to get into. Maybe we can give it a shot some other time.

To turn on the pump, I suggest we stick to what pHaestus presented in his article, here. I built it last week, and it's nice and simple. I added standoffs to the PCB, for a nice clean mount inside the case.

I agree with Yo_Duh: keeping the mobo in RESET is just sloppy. The mobo receives the +5 standby, and shorts the PWR_ON line (green?) to ground, which activates all the outputs of the PSU. The PSU sends a signal on the PWR_GOOD (or PWR_OK) line, as a +5v, TTL compatible signal, which the mobo receives and interprets as a ready state for it to start. All of this, including very tight timings is specified as part of the SSI standard ( www.ssi.com ? If it's still up.).

All we need to figure out, is how to tap into those lines. I think that pulling the pins out of the PSU cable would be simplest. We could run them through another Molex connector, for a nice/clean fit. Cost would be minimal. I can spec this out.

As for the soft shutdown, that's simple, using BladeRunner's pulse generator circuit, wired into the mobo's CPU fan speed sensor. Should we aim to allow a soft shutdown? Do we do it by delay, or by temp, or both?

Nice one!

Thanks for the FormFactor link too!

Compare "hacking a relay into a surge protector" to

http://www.overclockers.com/articles746/index03.asp

Asetek doesn't boot the system until temp reaches set point.

Was thinking that we should emulate the mfgrs. Seems only fair, as the opposite is the norm :)

is flow that redundant? if it was tuned properly (ok, this is probably not realistic) it would also keep the pump off if for some reason a leak/block/kink/other had developed?

Interesting idea.

The main purpose though, is making sure that a Pelt won't fry the CPU, and we're getting that by measuring the temp of the cold plate (aka spacer).

In order to add a provision to stop the pump (under those conditions), we'd have to add a differential pressure sensor, and I don't think we really want to go that route here.

I've been examining that option though, and it's possible.

Well direct flow measurement has the unfortunate drawback of compromising performance at least somewhat by lowering flow rates. So I too lean towards skipping it and relying on a temp. Will HAVE to get temp somewhere sensitive enough to changes that we have a nice safety margin though. Cold plate seems pretty close to ideal.

Agreed.

Here's a rough sketch of the logic:

i'll just throw in a few things around , hardware related :

Accurately reading temperatures with a thermistor or a resistor type probe is a very difficult thing. Most people dont realize how non linear , or time invariant it can actually be, specially with low temperatures, or very high ones. Measuring temperature in real time in a process is an actual pain in the @ss. Either with a thermocouple, thermistor or a Pt-X . They have a very slow response time due to the process of thermal transfer, and its variation. However you can have readings with less accuracy of more than 0.5 º Celcius that it wont matter that much. If it's 60 º or 60.5 º , it's not a big deal.

The level meter im trying to develop suffers the same. But because the variations are so small in the time frame , it's really not that important.

To deal with the first part, the linearity, you need to stabilize the readings by using a wheastone brige (preferably), reading the value between the two resistor divisors. Then you need to determine which voltage variation you want (Vin), and a correlation to white noise (you will have WN, but can be dealt with, more later). Lets assume WN is small , althou when you're trying to read a 1mV, it can be a nuisance . Then you need to linearize the readings into a more reliable signal, and more defined (and hopefully with again, low noise). So enter the OpAmp (non inverting). You will have some amout of troubles in the CMRR (Common Mode Rejection Rate) , but that can be regulated, by gain, and maybe an offset voltage in the inverting side, to give an ajustable interval of values in the output . Preferably an Instrumentation OpAmp setting, that can cut most of the noise factor.

So, lets (again) assume a somewhat linear output, with ajustment of offset . Ok, now you have a linear output of a temperature reading ready to be used in whatever you want.

Then you need to correlate the value read, to the value it's supposed to have to do X or W . Enter another OpAmp , a comparator with histherisis (err, dont know the correct word in english, sorry) , to switch at the given voltage . This will change the state it is on to a state that can feed a relay or an digital circuit (example) . It can go (lets say) from 0v (aproxim.) to , lets say 5v. The temp reaches the T1 value (a predetermined voltage) and it goes into High(5v). If the temp drops below T2 , it goes into Low (0v).

You can even use two, and a second temp reading from the hot side. Just a few ideias anyway.

PS: you can measure flow indirectly by several methods : with light refraction , magnetic displacement (if the liquid "cooperates"), Dopple effect or ultrasounds.

Yes my Digitecs use a wheatstone bridge. From a temperature monitoring standpoint, this is a great project because even 2-3C error is fine (whether a cold plate is cold or not is a pretty simple thing to measure).

hysteresis

Although this discussion seems to have abandoned the notion of using one, a flow sensor can be built using parts from a dead fan. I have built several. An example can be found here:

http://www.overclockers.com/tips642/index02.asp

The electronics are pretty straightforward, you just use a magnetic sensitive device called a Hall Sensor that latches "on" when a magnet passes by in one direction and "off" when the magnet comes back...

I have had a "fail-safe" circuit in my rig for a couple of years. If either the water flow stops or a the water block temp gets too high, the circuit trips a relay that leaves the ATX "power good" open until reset. I use the 5v line that is always on for the flow sensor and have a delay timer built in so the system can start with no water flow. However, if there is still no flow after 7 seconds, it shuts down.

Here is a schematic for a delay timer that operates within a certain period after power is turned "on". If you use it to control a relay switching output from the flow sensor, the fact that there is no flow when you first turn on your PC does not matter.

http://ourworld.compuserve.com/homep....htm#delay.gif

My thermal sensor circuit uses a comparator and is very simple. You don't need much when you are worrying about the kind of temperature swing that occurs with a cooling failure.

A schematic is here:

http://www.mitedu.freeserve.co.uk/Ci.../tempalarm.htm

Good to see you posting here again LMandrake: I have had that article bookmarked since it was written and whip it out for people sometimes :)


Good info with the comparators, I have a schematic SOMEWHERE for a temperature-based delay timer that BrianS designed using comparators.

Just a thought. Instead of having to monitor a temp, how about temp comparison. Say comparing the temp of the cold plate to the water temp. The cold plate has to be equal to or less than the water temp to pass the test. Wouldn't this be sufficient and simpler?

Not a bad idea, but it would involve a second probe.

Exactly what I was thinking. One on the cold plate, another on the incoming water to the WB, perhaps. Then a simple comparator. Could possible use comparisons of resistance using thermistors. Shouldn't be too hard to come up with a simple board for this. Quick, easy and cheap!!!

Should work. Your cold plate and water will both be at ambient at start up, so you would have to account for that. I wonder if there would be a few seconds where the cold plate might be warmer than the water temp right after start up.

----------------------------------------------------------------------------------
Should work. Your cold plate and water will both be at ambient at start up, so you would have to account for that. I wonder if there would be a few seconds where the cold plate might be warmer than the water temp right after start up.
-----------------------------------------------------------------------------------

Probably so, but the purpose is to not startup until the temp(resistance) is colder on the coldplate than the water temp. Been a long time since I've used my electronic skills (long, long time) but should be able to set it up so there's a set temp difference before allowing the setup. Like maybe the coldplate has to be 10 degrees colder than water temp before allowing start up. Don't think this would be too hard to design.

Wouldn't be hard at all if you used thermistors. Could just add a potentiometer inline with the thermistor for air or water temp and adjust the resistance so it was reading 10C lower than it actually is. Then when the comparator sees the coldplate temp as being equal to the other probe then the coldplate is actually 10C below water or air temp. I haven't messed with thermistors all that much, but I think this should work properly.