Peltier elements

an

e of a

ift

dB

True - they are more mechanically complicated. Even single stage Peltiers consist of lumps of bismuth telluride sandwiched between two layers of alumina. This doesn't make them slow.

My 1996 paper reports an initial 1.7sec exponential time constant on the cooling curve that preceded the 249, 414 and 589 second decay constants determined by the thermal mass of the block controlled and the thermal resistance to ambient.

The thermistor we used had a thermal time constant of about one second, so the lag inside the Peltier junctions - which were respectively 16cm^2. 9cm^2 and 4 cm^2 - can't be more than one second. That's not long.

We didn't see anything like that.

This doesn't make any sense to me. Thermal time constants are determined by mass of material and the geometry of its structure, and adding a heater isn't going to change that at all, nor move anything material closer to "the thing you care about".

It may move the operating point into a region where the transfer function of the Peltier junction - in terms of watts moved across the junction per amp driven through the junction - is more nearly stable, but that transfer function is known, calculable and predictable, and it's a lot more sensible to calculate the amount of heat you need to move and supply the current that you need to move that much heat rather than adding extra heat to be moved to make the calculation easier.

How does adding a heater make "the distances" shorter?

Controlling the heat transferred - rather than the current moving the heat - does let you tune the control loop to be dead-beat over the whole range, rather than forcing you to use a proportional term that is too low over most of the range to avoid instability at one end.

It's more computationally demanding, but thermostats don't need fast control loops, and the extra computation involved didn't come anywhere near overloading our micro-controller, in our application. Smaller, faster set-up could require more frequent up-dates, but there is a whole spectrum of cheap computational power available if you need something more potent than the Siemens SAB80C517A 8-bit microcontroller we used.

-- Bill Sloman, Nijmegen

of a

Sure, you can get 2 second-ish time constants on barefoot Peltiers, but that isn't necessarily fast enough. Normally you don't try slewing the temperatures of macroscopic objects that fast, because the gradients are hard to control. The butterfly-packaged laser I used in my downhole gizmo had a time constant of almost 20s barefoot, but a bit of local feedback a` la Phelan got that down to about 2s, which is where the uncontrolled phase shifts due to thermal diffusion took over. So time constants of that order with barefoot Peltiers are very doable.

Where I'm going with the Peltier-plus-heater scheme is rejection of ambient thermal forcing.

If your heater, sensor, and controlled volume are small, you win speed quadratically. A factor of 3.2 in size is enough for a factor 10 in speed, in situations where thermal diffusion dominates, and going from alumina to copper theoretically gets you another factor of 8.

Even just one factor of 10 gets you at least 20 dB better rejection of thermal forcing at all frequencies, and, more than that, it allows you to move the zero of your second order control law to a frequency 10x higher. That, in turn, means that frequencies below the zero benefit to the tune of 40 dB additional forcing rejection. That's a very big deal, since in many cases the diurnal or other low frequency forcing terms are by far the strongest.

Try turning a Peltier on abruptly at constant current, and watching what the cold plate temperature does. It overshoots and then recovers, and the phase shift of the recovery part is gigantic since it involves thermal diffusion in some millimetres of ceramic, which is slow, slow, slow.

Heaters on the cold plate are even more useful with stacked Peltiers, for all the same reasons except more so.

Only in the thermal-mass (read 'slow') limit. In that limit, the phase of the transfer function is a nearly constant -90 degrees on the rolloff portion, so you can always crank up the gain and get more bandwidth. (Slew rate is another issue, and depends on how hard you can drive the actuator.)

Once you're dominated by thermal diffusion, though, things get a lot nastier, because the phase shift increases without bound with frequency. (See e.g.

.) So small size is key, if you want a bandwidth greater than ~0.1 Hz.

It can be smaller, and closer to the volume of interest (e.g. a diode laser), and embedded in the same small piece of copper or aluminum, which diffuse heat about 6-8 times faster than alumina. A simple example is using the monitor PD in a diode laser as a temperature sensor. It's only a couple of mm away, and embedded in the same copper block, so it responds in a fraction of a second.

It isn't a control-loop problem that I'm talking about. You can't make a feedback loop work with thermal diffusion, because the frequency dependence isn't

H_LF(omega) = 1/( 1 + j*omega*tau )

which has a nice, well-behaved, asymptotically constant 90 degree phase lag, it's

H_HF(omega) = exp((1-i)*omega*L**2/Kappa),

in which every 1/e rolloff in amplitude gets you another radian's worth of phase shift. Once that sets in, you can't recover by tuning a PID--the only solution is to reduce the distance L over which heat diffuses. I go through the math of this in the chapter referenced above.

It's a completely different regime, but if you can make stuff work there, the forcing rejection improves amazingly.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net

coolers.

total.

I was going to make a fridge once. You have to worry about condensation and freezing possibility. That ice maker I have is neat, and makes ice pretty fast. Makes hollow cylinder pieces, that stay wet. The ice builds up in the storage, and keeps melting. That reduces the total refrigeration needed, and keeps the ice from sticking together.

I always had this thing at work on display. 2 foot copper probe, 1/2 inch thick, with heavy base. Must weigh 5 lbs. Experiment gone bad. They were attaching peltier devices on the base to cool the rod. I forget what the probe was for, but I don't want to get into that. Ridiculous.

I've worked with tons of microscope heater/cooler stages. Newer ones use peltier for that. Some use water tubes for heat transfer. Switching supplies no good for things that need low electrical noise. The control voltage is a slow ramp DC. Going positive or negative. Usually the small working area is small, so external sinking is minimal

Greg

with an

ponse of a

o shift

. 20 dB

a

ust

rly

use

Which I'd describe as feed-forward control. If you know what's happening on the exhaust - heat-sink - side of the Peltier junction, your scheme can change the amount of heat being generated in the controlled volume in such a way that the amount of heat that needs to be transported through the Peltier junction remains unchanged. You then need to know the temperature of the heat-sink almost as accurately as you know the temperature of the controlled volume - the thermal inertia of the controlled volume and the integral term of your PID control loop mean that you don't need to know the temperature of the heat-sink quite as accurately as you know the temperature of the controlled volume - but that's just making the circuit marginally more expensive.

o , *2

rs,

That's pretty much what we did to measure the time constant of the controlled block plus Peltier junction - though we did use turn-off rather than turn-on - and what we saw was monotonic - a quick small exponential decay with a time constant of about 1.7 seconds followed by the much slower exponential of the block cooling as a whole.

Heater plus feed-forward control.

me

.

And a temperature sensor that's got low thermal mass and is in really good thermal contact with volume whose temperature you are trying to control, which is to say a low thermal time constant.

2. >

Okay. It's a separate local control loop, inside the the larger control loop that uses the Peltier junction to shift the bulk of the heat and generate the bulk of the temperature difference from ambient.

Basically another version of your two-zone heat temperature control scheme

No, it sounds more like two essentially independent nested control loops.

But what you are now describing sounds more like two nested control loops, the outer one doing the heavy lifting, and the inner one providing the agility.

To change the subject a bit, one scheme I've come across in precision calorimetry uses a Peltier junction as the sensor in a fast local AC- only local control loop.

The Peltier - okay, Seebeck in this application - junction is sandwiched between the calorimeter and a solid, well-insulated block of copper - a thermal capacitor - which eventually thermally equilibrates with the calorimeter, so the Seebeck junction only monitors the transient temperature differences.

The advantage of this approach is that the Johnson noise on the output of the Seebeck junction is much lower than that of a platinum resistance sensor and the sensitivity - in volts per degree - is very much higher.

You are still reliant on some other temperature sensor to keep the controlled temperature stable in the long term, but the short term stability can be improved - if, of course, only up to the limit set by thermal diffusion times.

-- Bill Sloman, Nijmegen

with an

ponse of a

o shift

. 20 dB

a

ust

rly

use

o , *2

rs,

me

.

2. >

t -

Wow! That's great! The next project looks to be on the 'slow road to china' for exactly the above reason. But it's hard to get people to understand. It's 'just' a bit longer time constant isn't it.. (I may copy and paste that, perhaps it will mean more coming from your mouth.)

I sometimes think I'd like to build a thermal test-bed 'thing'. Different sensors, 'plants', hunks of metal of various lengths, loads ... Hook it up and control the T, and then beat on it in different ways. I'd learn a lot!

I think I 'get' the TEC + heater idea. It's a bit specialized, but you keep the TEC cooling (a heavy idle.) and then control things short term with the heater. It's a 'class A' brute force solution... Just my style :^)

Bill, you should give Phelans book a read. I got a copy for ~$15, delivered.

Speaking of thermal diffusivity, I saw an ad for graphite thermal sheets.

George H.

an

response of a

shift

dB

No, Bill, it's just plain feedback. You just get way wider bandwidth, so the loop gain is higher at any frequency below the unity gain cross. This really isn't a hard concept.

If you know what's

Nonsense. It's just ordinary feedback, it just has a much faster actuator. No mysteries.

No, it isn't a polynomial rolloff at all at higher frequency. Do the math if you don't believe me.

No, once again, no feedforward. You don't have to measure the sink temperature at all, and it wouldn't help you if you did.

The sensor and the controlled item have to be small, sure. I said that already, I believe. But once again, unless you have a fast actuator,

*it isn't a time constant problem at all*, because the math is different once diffusion starts to dominate.

No, you only need one sensor per controlled zone. It would be quite possible to have multiple zones on the same cold plate if you needed to, each with its own heater. You'd have to manage their mutual interactions, but that wouldn't be too hard if the bandwidth were large enough.

No, once again, it's a fundamental speed limit due to the onset of diffusion-limited propagation.

I think I've been adequately clear. The chapter I posted has a lot more detail.

Sounds potentially reasonable, but if you can insulate the copper block well enough that ambient forcing doesn't move it around, why not just do the same thing to the controlled volume and save trouble? Is it mainly for dealing with variable dissipation in the controlled zone?

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net

or with an

response of a

is to shift

i.e. 20 dB

delta

exhaust

cularly

ven use

f

wo

n d

ut

the

are

e

al

ime

Then it's one feedback loop controlling two actuators.

How do you partition the feedback between the slower Peltier junction doing the grunt work, and the little local heater providing the agility? Does the Peltier junction just get the integral term from the PID output?

I don't want to try and re-invent that particular wheel - if you were prepared to be more explicit about what you do, it would be a waste of effort and bandwidth.

Apart from the unspecified way you managed to develop to partition the output of the control loop into a control signal for the Peltier junction and a control signal for the local heater.

d n m f u

it to

deal,

are

nd.

e I**2

e bars,

hat

and

,

One could do a more or less infinite amount of math on a more or less realistic physical model of the thermal mass being controlled, including the thermal lag from the mass to the sensor, the extra thermal lag from the local heater back to the controlled thermal mass, and the different - and somewhat larger - thermal lag from the Peltier junction back to the controlled thermal mass. For extra credit you could throw in the thermal mas of the heat sink on the exhaust side of the Peltier.

Without real experimental feedback to test your choice of the simplifications required to keep the model tractable, it wouldn't be a particularly useful activity.

Since I never saw anything that vaguely looked like your weird overshoot, I'd probably be doing the math on a rather different system.

Of course it could help - the effectiveness of the Peltier junction depends on the temperature difference across it, and if you don't know that temperature difference you can't fully optimise your control loop.

I can accept that you don't bother calculating exactly what the Peltier is doing for you, though I would see it as a missed opportunity, but until you tell us how you partition the feedback between the Peltier and the local heater you are essentially saying that you have a magically better way of managing your system which you aren't going to tell us about in enough detail to let us copy it.

ved

e same

t's

bout.

nd

g

hase

off

e

.pdf

Hz.

The control theory books that I've looked at distinguish between controlling an inertial mass and and controlling an inertial mass with a time delay in the feedback loop. The math for these two conditions is - obviously - different, but the texts did point out that time delay was always present in any real system.

he

e,

d o

es

of 10.

r

That I can see. but I am still interested in how you partition the feedback signal between the Peltier junction and the local heater.

he

t t

ere

True, but you are being a little mysterious about the details of the control loop that you use to get around it.

Or - in control theory language - the presence of a fixed delay around the feedback loop. Because the delay through the heater is less than the delay through the Peltier junction, you've got two different feedback loops in the one system, and you haven't told us how you split the error-correction signal to exploit these two separate actuators with their two different sets of dynamic properties.

While your inner heater doesn't have as large a pure delay as the Peltier junction, it is still in series with the temperature sensor, whose pure delay forms part of the pure delay around both of the control loops. Feedforward does have the advantage that it by-passes the delay in the sensor (but not the actuator), but it is a purely open-loop correction.

se

th

I'd beg to differ. You haven't been at all clear on how you contrive to exploit the different properties of the two different actuators - and my recent posts make my incomprehension perfectly obvious.

Spamming your book? Have sales been slow recently? And does that chapter spell out how to partition the feedback signal between the two actuators?

If it does I'll probably buy the book (but not until I'm safely in Australia and can stick it on the shelf next to "The Art of Electroncis" and Williams and Taylor on filter design, that are out there already.

As I said, I came across it in a paper on precision calorimetry, which is - of course - all about measuring the heat that gets dissipated on the controlled volume aka calorimeter.

-- Bill Sloman, Nijmegen

If not explained in any kind of useful detail.

I've not yet got a text on control theory. I've dipped into quite a few of them, and got what I've needed, but the ones that I saw confirmed Sturgeon's Law, that 90% of everything is rubbish. If Phelan's book is part of the 10%, it's quite likely that I'll buy it.

Used it back in 1993, so it's referred to the 1996 paper. It's electrically conductive, which can be inconvenient with power transistors, but it's great for power resistors aka heaters and Peltier junctions. Unlike Thermopads, you did seem to get the thermal conductivity that the manufacturer predicted.

-- Bill Sloman, Nijmegen

with an

response of a

to shift

20 dB

sigma-delta

exhaust

particularly

use

I**2

bars,

same

about.

2. > >

Not exactly. The phase does go linearly with frequency, but unlike a normal delay, the amplitude also falls off exponentially. Effectively the at any given frequency, the loop loses sight of anything more than a few diffusion lengths into the material.

The problem with feedforward is that it has to be very accurate, or it limits the forcing rejection rather than improving it. Insulation and feedback don't have that problem. Which is not to say that feedforward isn't useful in some situations, it's just not what I was talking about.

You can do it that way, or alternatively you can run the Peltier at a fixed current and rely on the heater to do all of the work. Arranging the loop so that at zero error you're dissipating 1/4 of maximum heating will allow both actuators space to work in without having two control loops fighting.

From other folks' comments, they seem to have understood. We all start from different places, of course, and there's no dishonour in that.

That chapter is a free download, and speaks exactly to the point at issue.

It would be nice to be able to argue without quarrelling.

Read it in good health.

Makes sense.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510
845-480-2058

hobbs at electrooptical dot net
http://electrooptical.net

y

phase

worth

a t

ced

k

What's magic about a quarter? I imagine you mean that the close-in resistive heater is dissipating 25% of the maximum power you can afford to have it dissipating. I'd have though that that running it at half the maximum dissipation you think that you can get away with would maximise your headroom, but obviously there may be subtleties in there that I don't know about.

They may not have known enough about the gritty details of putting a temperature controller together to have actually noticed that you were being vague and elliptical and skating over important detail, or - more likely - weren't interested enough to care. Obviously, there's no dishonour in being vague and elliptical - it saves a lot of typing, and if you been immersed in a particular technology for a while one does tend to slide over the practical details which have been drummed into your head by repeated exposure.

That would be Chapter 19 which didn't actually make it into your book

- which could be just as well. The phrase "carbon thermistors" came up twice, and thermistors aren't any kind of carbon resistors but a sintered mix of metal oxides or sulphides. Faraday apparently made the first one out of silver sulphide.

I couldn't find an answer in it to the question of how you partition your control signal between the fast local heater and the slower Peltier junction - which is not to say that it isn't a brilliant and totally admirable bit of work. The "carbon thermistor" pratfall doesn't detract from that at all - you are perfectly entitled to the occasional human error, and - if pressed - you can always claim that your publisher insisted that you included a couple of deliberate - and obvious - errors as anti-piracy devices.

Who is quarrelling? I'm trying to get you to be more explicit about what you actually do, and I'm getting a bit grumpy about the time that it's taking to define exactly what it is that you are actually doing, but I certainly don't hold you in lower esteem now now than I did when this thread first got going.

I'm about to turn 70. My shiny new aortic valve is working fine, but the other heart valves are a bit leaky - nowhere near enough to create any kind of problem, but I may see more of my Australian cardiologist than I like (and he's a very pleasant character, if totally brilliant in his field - one of my friends from undergraduate days is a professor of oncology with a dodgy heart, and he got his cardiologist to take me on too).

I'll do my best to follow your instruction, but I can't promise to succeed.

-- Bill Sloman, Nijmegen

a

If I was to pull a solution out of thin air, I would run the Peltier so that the resistive heater ran at half power on average, thus giving me equal dynamic range in both directions. That's probably close enough for government work. For extra credit, I would shift my operating point just a tad to compensate for ambient heating (or cooling - whatever the case may be).

Oh, I think I know how to do that. (But I've never done it for a thermal thing.) If you've got two actuators and one (processed) error signal, then you can split the error signal (frequency-wise) and send the DC part to the slow actuator and the high frequency AC part to the other.

You can do a similar trick for locking an (external cavity grating feedback) diode laser to some absorption line. The grating position is controlled by a (slow) piezo and then you can do a fast tweak with the laser current.

Now a laser current can go both ways, not so with Phil's little heater. So I'm assuming you've got to add a bit of bias to the peltier.... Whcih means some of the DC signal also has to go to the heater. Getting the balance of the frequency split and the gain vs freq for each arm is where you 'get paid'.

Oh... (sorry thinking out loud at the keyboard) Forget some of that above. You can just run a little bias through the heater and then the peltier will take care of it! That will give you enough room to turn the heater up and down.

Cute trick if you ever need fast T control.

George H.

1, 6:14=A0am, Phil Hobbs

It is. It took me a while to work out what Phil was doing and why, but now that I've got my head around it, I'm impressed.

In the system I put together, only about half the "pure delay" around the control loop came from the Peltier junction - the lag within the thermistor must have been just as big. To get the maximum advantage from Phil's approach in that system you'd want a rather faster temperature sensor.

If Phil is mainly thinking in terms of multi-stage Peltier junctions, where the thermal diffusion through successive stages will add up to a rather longer delay than I saw in my single-stage junction, he'd probably see quite a lot of advantage with even a relatively ordinary temperature sensor.

-- Bill Sloman, Nijmegen

get the result I was looking for. I wonder how they make those mobile coolers.

enough to set of the heat - it got quite hot, 90 deg. C, and with a fan some

45C. That means, that the "cold" side was still warmer the otherwise inside temperature (26C at work, at home 21-22).

surprised then instead if getting a cooled part I actually got a heater in total.

to have a larger and more effective part for getting rid of the heat.

you need giant heatsinks on the hot side, sheet metal from a CD player won't work. Large CPU heatsinks are ok, but the bigger the better, and most fans from CPUs are too small for the typical 1.5"-ish square units out there. The bigger the heatink and cooling on the hot side the better.

I've not done this yet.

21, 6:14=A0am, Phil Hobbs e

seems to me that it is real similar to a class-a single ended amplifier

the peltier as a constant current(heat)sink, the heater as a variable current(heat) source

-Lasse

If you are trying to move 5 watts of heat, a water cooled heatsink is essential with Peltier.

--
Many thanks,

Don Lancaster                          voice phone: (928)428-4073
Synergetics   3860 West First Street   Box 809 Thatcher, AZ 85552
rss: http://www.tinaja.com/whtnu.xml   email: don@tinaja.com

Please visit my GURU's LAIR web site at http://www.tinaja.com

=A0units

r.

There are other ways of moving a lot of heat - heat pipes worked for us back in 1994. Admittedly, these are "water cooled" but it's evaporative cooling in a sealed system. They replaced the original water-cooled system, that gave as trouble - in the long term - when air-bubbles started appearing inside the pipe-work (presumably due to diffusion in and out of the water loop through the plastic tubing we used to link up everything - though the pump seals probably weren't all that perfect either).

-- Bill Sloman, Nijmegen

What are you assuming for power input? Temperature difference? Ambient?

With plausible assumptions and looking at some typical curves, it looks like 25W might do it for my hypothetical situation.. so 30W at the hot side, which doesn't look to be much of a problem for a fan-cooled heatsink, but a bit expensive and big for a passive heat sink. Modern CPU coolers should be many times better than required (even though they might add another 5W for the fan).

Is there something I'm missing, Don? I assumed 20°C controlled temperature and 50°C maximum ambient.

Of course if you're trying to make a refrigerator or freezer that will operate in the summer in AZ the numbers get a lot less pleasant. Some years ago we made a nice little milk cooler as a part of controls for a commercial capuccino maker and it worked fine with a fan-cooled heatsink in an office or restaurant type of environment.

Best regards, Spehro Pefhany

--
"it's the network..."                          "The Journey is the reward"
speff@interlog.com             Info for manufacturers: http://www.trexon.com
Embedded software/hardware/analog  Info for designers:  http://www.speff.com

I just saw some peltier junction milk and creamer coolers at a coffee shop here in Chicago.

I could see the heatsinks through the rear of the cases.

Of course one wasn't even plugged in. The one that worked had some weird milk pump in it for dispensing. I assume it never gets flushed and cleaned.