VariablePC Peltier Controller

[koslov]

Here's a cool new product for TEC users: http://www.techcool.com/vpc/photo.htm

Controls the power consumption of TECs based on temps of cooled object, using PWM to vary power and up to two thermistors to measure temps.

Looks very neat, but expensive. Also, the highest rated unit can only handle a 172W pelt. But none of the units are capable of supplying Vmax for 24V TECs. The best they can do is 18V. I would have preferred if the voltage was adjustable, so you can operate the TEC at full power if needed. For 15V pelts, you have to use the 12V model, 3V under Vmax.

What I find a little funny is this quote:

What it doesn't do. . .
Waste power - the VPC controller supplies only the amount of power necessary to do the work (cooling). It does not take a higher amount of power from the power supply and reduce it, rather it only takes the amount necessary, allowing for cooler computer operation.

This seems to imply that no power is wasted, or at least very little. Well, what is that big, honkin' heatsink for on the larger models??

What do you think? Overpriced? Can a DIYer build one significantly cheaper?

[Since87]

Originally posted by koslov
Looks very neat, but expensive. Also, the highest rated unit can only handle a 172W pelt. But none of the units are capable of supplying Vmax for 24V TECs. The best they can do is 18V. I would have preferred if the voltage was adjustable, so you can operate the TEC at full power if needed. For 15V pelts, you have to use the 12V model, 3V under Vmax.

It looks to me like the VPC-225+ is spec'd to handle a 226 Watt TEC at Vmax. The website is ambiguous though. The VPC-225+ may be vaporware.

Originally posted by koslov

What I find a little funny is this quote:

What it doesn't do. . .
Waste power - the VPC controller supplies only the amount of power necessary to do the work (cooling). It does not take a higher amount of power from the power supply and reduce it, rather it only takes the amount necessary, allowing for cooler computer operation.


This seems to imply that no power is wasted, or at least very little. Well, what is that big, honkin' heatsink for on the larger models??


Suppose power to a 226 Watt TEC is being provided through this thing at Vmax, that's 365 Watts controlled by the VPC-225+. If the VPC itself is dissipating 10 Watts, I'd consider that a very small amount relatively speaking. With only passive convection to count on for cooling, that heatsink looks reasonable to me.

Originally posted by koslov
[B]
What do you think? Overpriced? Can a DIYer build one significantly cheaper?

At $106 + shipping, I'd say the price was 'reasonable'.

It would certainly be possible for some to build an equivalent device cheaper. Particularly if packaging it nicely were not an issue. I think most would have a difficult time doing it for less than the VPC price though.

The biggest problem I see is the way it operates. My understanding, (based on an article at Overclockers some time ago) is that these things cycle the power at a very slow rate. (A few seconds on, a few seconds off.) TEC manufacturers recommend against this type of control. In fact the manufacturers recommend keeping the 120 Hz ripple of 'DC supplies' below 10%.

The reason the manufacturers recommend against ripple and turning the supply on and off at a slow rate, is that these conditions age the TEC prematurely due to excessive temperature cycling. Personally, I'd avoid setups that could lead to early TEC death.

[koslov]

I found the overclockers.com article: http://overclockers.com/articles424/

I thought the wattage rating was output. At 12.6V, the 226W TEC would consume ~250W.

Here's some more info on the different types of TEC control: http://www.marlow.com/TechnicalInfo/power_supplies.htm

VariablePC claims,
Dual stage Pulse-Width-Modulated (PWM) power
low stage power is constantly supplied to the controlled TED
high stage power is automatically supplied to maintain the selected operating temperature
reduced thermal cycling helps maintain TED's reliability
From this I would deduce that they are not using a simple on/off controller.

[Geoff]

From my standpoint I'd say $106 US is lot in terms of what the parts cost. However they probably have a custom programmed micro doing both the pwm and pi control of the temperature. And that probably cost them quite a lot, especially if they sourced it out. So they are just charging for the R&D. Then 106 dollars is perfectly reasonable to anyone who can't do that sort of thing for themselves.

Since87 is right about the switching, I could only imagine all the horrible harmonics that pwm might produce, but a simple second order low pass filter (inductor and capacitor) should be able to take out most of the harmonics and leave the dc component behind so you have nice more or less clean output.

Personally I would like to try this out myself and don't think it should be too hard. The pwm is easy to program in a micro, but the PI control, that will be a little harder since I've never done it. But one of my textbooks outlines a few procedures in implementing PI and even PID control in a micro. So once I have a chance I'm going to start.

[Since87]

Quoting the overclockers article:

"What I found is that temps cycled over a 15 C range until they settled into a 5 C range, after about 5 minutes. The "Working" led settled into a pattern of 5-6 seconds ON, 8-9 seconds OFF."

That sounds fairly bad to me.

I can't say for certain, but I doubt there's anything remotely approaching the sophistication of a microcontroller running a PI routine in these things. My guess is some comparators and a bit of discrete logic.

Still, the parts cost to do this as a 'one-up' is going to be fairly high for a lot of people.

[Geoff]

oops, Its been a long day and I confused two products.

I thought we were talking about this

http://mcshaneinc.com/html/5C7-350.html

Its a little cheaper then that vpc thing, can be daisy chained together and it actually uses PI control, although I've been told that PID is better for heating/cooling processes.

[koslov]

Here's some questions I sent to info@variablepc.com:
Hello,

What kind of control logic do your VPC controllers use? Is it one of these: http://www.marlow.com/TechnicalInfo/power_supplies.htm ?

What have you changed in your design since it was reviewed here: http://overclockers.com/articles424/ ?

What is the highest recommended TEC model for your 225+ unit?

Thank you,

Koslov

And the reply:

Hi Koslov,
It is a modified proportional control.
The circuit originally was an on/off circuit...so that has changed.
The pot is different and allows for a wider range of adjustment.
There is a 'remote' available (180r, 200r, 225+) that allows you to mount the controls and leds away from the head unit.
The passive heat sinks are larger.

If you drive a 226watt peltier with the 225+ there is some 'head room' but I would not suggest trying to drive much more. Actively (fans) cooling the heat sink will add some stability (linearity) to the temp control. The driving mosfets are under-rated by 20% but the heat associated with controlling a 226 watt peltier is the most significant obstacle.

jim

So the model rating is for the max TEC rating, not output (odd that it clearly says "Output: 225WATTS-max" on the website). If you can adjust the voltages with the pot, I guess the 12.6 and 18 given are just nominal.

The McShane controllers look nice, but lack the dual probes if you are trying to keep that CPU at ambient. Not a problem if you're ambient temp doesn't vary greatly, but at least mine does. If you got the 5C7-350A kit ($62), you'd still need a probe, around $10. But they do have a probe that's < 1mm diameter! Then you need heatsinking and an enclosure perhaps. So it's definitely cheaper to DIY, but the savings aren't that big.

Hmm, both of these units look nice, but it'd be nice to see how well they work in practice. A review would be nice, but do you know any pHellows with temp equip?;)

[Geoff]

It is a modified proportional control.

So apparently its just proportional control. I wouldn't use it though, with proportional control there will always be steady state error ( once everything has settled down there will still be a noticable difference between setpoint and output ). So if you set the thing to be at 5 degrees C (just above dewpoint?) then you don't know with any certainty how much the error will be it might be small like .5 degrees but it might be large like 5.

PI control is by far much better as it introduces an integral mode into the formula. Integration is basically the area under a function between two points, so in control terms it pretty much takes the sum of the error over a change in time, therefore so long as an error exsists the integral mode will continue to increase its output while the proportional just sits there and does nothing. Plus with PI control the controller gain can be raised much higher, before instability occurs (compared to porportional control alone) allowing for a much faster responce time, without being un-necessarily underdamped.

Personally I would use proportional control on simple things that don't require a great deal of output accuracy, but in situations where a decent responce and output accuracy are concerned PI control is much better. So in the application of TEC control, a controller with PI control would be better because of its increased responce time and zero steady state error. So the VPC would definately be a no go especially if a TEC controller with PI control and a fast PWM frequency with properly filtered output is available.

Here's another link its one of maxim's design notes on TEC control, it would have to be modified for our application though. The H-bridge allows for heating and thats not good, plus the mosfets would have to be change to handle the current of higher powered TEC's. It looks fairly straight forward with lots to wire up.
http://www.maxim-ic.com/appnotes.cfm/appnote_number/1757

[Since87]

Geoff,

If you come across any good websites explaining how to implement a PI controller in software, let me know. (I haven't looked.)

Off the top of my head it seems fairly simple, but if there are tricks to make it simpler, I'll happily use them.

[Geoff]

Since87

I have one textbook ( at school right now ) and it more or less outlines a basic algorithm for doing it. Along with a sequential flow chart, or at least what looks like one.

If I can remember to scan those pages, I ll post them up, but I got a term test, final project to work on, lan party to go, fixed mobo to return and help a friend move in to his new place for the weekend, plus squeeze the gf in there somewhere. So there a good chance I ll forget.

[Geoff]

Since87

I looked through the book and the flowcharts seem to explain things the best so here they are.

The pic shows all three modes P, I and D. The charts work on the basis that all three are working in parallel off the same process variable. I don't know how much you know about control theory so I ll explain the charts just a bit.

P mode -
DE = error, DV = process variable, DSP = setpoint, CORR = proportional correction, KP = proportional gain, P = proportional output, P0 = output bias

I mode -
DE, DV and DSP are the same as from P mode, sum = sum of errors, PI = integral output, KI = integral gain, DT = change in time between error samples.

D mode -
DE, DV and DSP are the same as before, DDE = the difference between current error sample and previous error sample, DEO = previous error sample, PD = derivative output, KD = derivative gain.

So If you want PI control all you do is implement the P and I flowcharts add the results together (controller output = P + PI) and there you go. For PID add all three. Theres two problems that I see might occur with this, too big of a DT and the rectangular integral method gets inaccurate. Too small of a DT and your program won't crunch the numbers out fast enough.

[Since87]

Thanks.

Originally posted by Geoff
Theres two problems that I see might occur with this, too big of a DT and the rectangular integral method gets inaccurate. Too small of a DT and your program won't crunch the numbers out fast enough.

I'd guess that with a TEC cooled CPU, a 1KHz sampling rate would give both a small enough dT to give an accurate enough integral, and plenty of time for most microcontrollers you'd want to use to get the calcs done. (I'm assuming the input signal is coming from a sensor on the coldplate and not directly on the CPU. I'm guessing that the copper of the coldplate, 'low pass filters' the fast changes in power dissipation that may be going on in the CPU.) There's a lot to learn in trying to really fine tune a system like this.

I'm looking at getting my employer to buy me an Altera Cyclone FPGA development board that can run a 'soft' 32 bit processor at about 100 MIPS. I don't expect processing power to be an issue. (Major overkill I know, but it would give me a chance to learn the Altera tools and learn my way around the 'soft' processor which we are considering using at work. I'm doing mostly Xilinx development lately.)

[Geoff]

Tuning shouldn't be problem though, I was thinking you set some of the gain settings to really fairly conservative ones, then use whats called "process reaction curve method" to then properly figure out the values. Or try to get this piece of software and get it working.
http://bestune.50megs.com/

It looks interesting and apparently gives really good results plus the interface looks really simple as compare to a package like Matlab.

Personally we didn't cover that much FPGA's in class, we played around with some Xilinx dev boards nothing serious.