The Heat Die

[bigben2k]

As part of a test bench, we need a steady source of heat, that simulates what a CPU puts out. A CPU is a poor choice here, because the heat level is difficult to control, and to measure.

The heat die may use electrical heaters, the output of which can easily be measured by monitoring the voltage and amperage going through it. The accuracy of this measurement will be discussed here so this thread may be more about this heat source, its construction, and very importantly, how to prevent or minimize the secondary losses, which may throw off results.

Temperature measurements at the heat die may be discussed here.

Some references:
-Joe's sim (Is there a link?)
-Bill's bench
-RoboTech's page

[nicozeg]

Originally posted by bigben2k

-Joe's sim (Is there a link?)

Yes!
Also I remember some forum thread with pictures of bill's simulator, but I dont have the link. (it was'nt on procooling)

[joemac]

My suggestion would be to use a T.E.C as a thermal source. A unit like this should generate a constant temperature needed to test a block providing power is adequately regulated.

[MadDogMe]

A TEC sounds dodgy, it's not constant. Would'nt it's 'performance' vary with the coldside temp, which would vary with the hotside temp. To much room for fluctuation?.

One of those big resistors they use for discharging battery packs would be better suited I think. I seem to remember Billa suggesting something better though?(surprise, surprise!)...

[Boli]

A soldering iron.

I don't know if they work on a DC current but if they do then you could butcher an old one up replace the tip with a heat die shaped one, and give it current through a circuit with capsitors (to regulate) and resistors (to alter the current).

They CAN get up to VERY hot temperatures and run for hours (well at least the ones I have do... I'm allways leaving them on.)

Worth looking into?

~ Boli

[bigben2k]

Actually, I was refering to Joe Kelly's bench, not Joe Citarella, but that's a good link, thanks!


joemac: I understand, from correspondence with your partner, that you've been looking at TECs because of this issue that resistor's actual values can shift over time and use. My reply to that (someone correct me if I'm wrong) is that it's really irrelevant, because regardless of the actual resistance, the power is measured by the voltage, and amperage.

So I'd refer you to RoboTech's heaters (if I had a link). They're that small cylinder that you know about. They can be mounted inside of a copper or brass die, and then insulated, to minimize the secondary heat losses (a source of error, at ~2%). A power resistor may work, but it may be a bit less convenient to deal with, I think. Dan From dansdata.com uses one.

As for TECs, I don't know how reliable that would be, as a heat source: Since it "moves" heat, using the power that is provided to it, the end result may be somewhat unpredictable. You also have to consider the clamping pressure, which could then become another variable, and source of error. We want to minimize the errors, as much as possible, because we can barely afford it.

So a heater/resistor is easier to deal with, because the power measurement is direct, and the accuracy is simply down to that of the meters.

One thing I've been pondering on and off for some time, is if it's possible to use the same meter, and switch back and forth, between measuring amps and volts. Then it occured to me that the amp measurement, if it's a tap in the power line really can't be interrupted, but there's always the amp probe, or alternatively, if it's acceptable, a power shunt, which has its own set of variables... But that's a discussion for this thread.

[joemac]

Ok let me explain this better and maybe a better drawing will help. The cold plate is not just a plain pressure plate as I showed in the original drawing. It is a heat sink enclosed in an insulated environment. This will prevent changes in the temperature outside to affect the temperature at the simulation die.

As for the resistor if W = VI and I = V/R a change in R (resistance) would cause a change in “I” this will cause “W” to change. A T.E.C is less prone to “R” changes if used and installed properly.

[bigben2k]

joemac: you may be right in that the resistance of a TEC doesn't vary, but what I was trying to say, is that it's irrelevant: we can still measure the power output of a resistor/heater, by multiplying the amps and volts going to it. Simple.

Yes, the resistance of a heater/resistor changes, as you apply more power to it. Other than than, maybe some vibrations might change it, during the test, but that would still show up in the voltage and amp measurements.


As for the TEC, how would you know, for sure, that you have the same exact clamping pressure, from one day to the next? Are you going to "torque" the coldplate on it, every single time? What kind of errors do you think that would give you? How could you measure that error?

Also, you still don't know how much heat the pelt is outputting through its hot side. Although it's proportional to the power that you give it, it's not "directly" proportional. So how would you measure the power?

A tec is great if you're trying to get to a specific temperature, but that's not the case here: we want to get to a specific power output.

I like your diagram. You'd have to extend the insulation, to make sure that the cold air doesn't find it's way back up. If you use metallic screws, some of the output is going to end up being dissipated on the cold side. So you'd still be back to measuring it.

[Groth]

Here's a crudely drawn representation of my heat die concept. A cartridge heater (shown as red, such as the ones pH has - Watlow Electric?) is imbedded in a chunk of copper. Heat flows up the riser towards the water block. Two thermocouples are embedded into the riser at, say 10 mm intervals.

At steady state the temperature gradient up the riser will be linear and proportionate to the heat flow. Since we know the cross sectional area of the riser and the distance between the thermocouples, we can calculate the heat flow from the temperature difference. Additionally, we can infer the temperature of the die surface without having a probe directly on the die, avoiding the problems of a 'heat shadow'.

[iggiebee]

Resistor assy used by Dan's Data here

[jaydee]

Here is the slug I built with the round watlow heater cartride.



Note there is 2 "steps" on both sides below the die area. Those are for TC's on both sides. You do not want to put them into the main raised part as that will effect the heat transfer to the top.

[nicozeg]

Originally posted by Groth
Here's a crudely drawn representation of my heat die concept. A cartridge heater (shown as red, such as the ones pH has - Watlow Electric?) is imbedded in a chunk of copper. Heat flows up the riser towards the water block. Two thermocouples are embedded into the riser at, say 10 mm intervals.

At steady state the temperature gradient up the riser will be linear and proportionate to the heat flow. Since we know the cross sectional area of the riser and the distance between the thermocouples, we can calculate the heat flow from the temperature difference. Additionally, we can infer the temperature of the die surface without having a probe directly on the die, avoiding the problems of a 'heat shadow'.

Thats a very smart idea, a way to measure power without the need of amp and volt monitoring. It has the advantage that it counts only the heat flowing to the die, independent of the possible secondary heat loses.

I'm too lazy to put some numbers on this right now, but the critical factor is going to be the temp probe accuracy, as the temp difference to measure can be small. But precicion of the readings can be increased with probes separation.

[gone_fishin]

Originally posted by Groth
Here's a crudely drawn representation of my heat die concept. A cartridge heater (shown as red, such as the ones pH has - Watlow Electric?) is imbedded in a chunk of copper. Heat flows up the riser towards the water block. Two thermocouples are embedded into the riser at, say 10 mm intervals.

At steady state the temperature gradient up the riser will be linear and proportionate to the heat flow. Since we know the cross sectional area of the riser and the distance between the thermocouples, we can calculate the heat flow from the temperature difference. Additionally, we can infer the temperature of the die surface without having a probe directly on the die, avoiding the problems of a 'heat shadow'.

You are going to infer the temp? At what resolution? As an example will it be 70C or 70.1C or 70.01C? Even if you place sensors in the riser as you plan there is still the calibrated accuracy of the probes to be figured in. As an example even with equiptment capable of .01C resolution, there is a +-.05C accuracy to be figured in, doubling the probes doubles this problem.

[sevisehda]

Groth: That is one brilliant idea. It may be too perfect though, at times its difficult to make the math fit the real world and this may be one of those times. Most likely it would would but it would need to be calibrated quite accurately.

BigBen: Why not place a resister between your PS and load, then measure the voltage drop across the resister. Then calculatte the amperage by V=IR.

[bigben2k]

That's an interesting concept, Groth!

I would think that in order to minimize the secondary losses, one would want to keep the heater as close as possible to the top of the die. You may have just gotten rid of that issue, to some extend.

So does anyone know what the space shuttle uses for thermal insulation, and if there's anything similar we can use?

[nicozeg]

You are going to infer the temp? At what resolution? As an example will it be 70C or 70.1C or 70.01C? Even if you place sensors in the riser as you plan there is still the calibrated accuracy of the probes to be figured in. As an example even with equiptment capable of .01C resolution, there is a +-.05C accuracy to be figured in, doubling the probes doubles this problem.

It's not difficult to calculate. Lets start with round numbers:

-A heat die of 100mm2
-100 Watts load

That makes a heat density of 1w/mm2

With copper conductivity of 0.386W/mm*k we have a gradient of 2.59ºC for each mm in lenght. With probes 10mm apart there is a delta of 25.9º. If you use two probes with an accuracy of 0.1ºCfor an accumulated error of 0.2º you have a power measurement accuracy of 0.77%. That's not bad at all considering that when monitoring input power only you have to guess how much is your secondary heat loss.

Calibration of the two probes is easy: in cold and steady state both should measure the same temp.

Applying this calculations to Robotech's die simulator we we can see a real world case:

Asume again he's using 1w/mm2 heat density. Given that he's using a brass die (I'll guess 0.120w/mm*k?), the gradient would be 8.3º/mm. In a distance of 10mm the delta is 83.3º. He's using probes with 0.05º accuracy so his error margin is 0.12%.

By choosing brass for the heat die the gradient have a big increase and improves accuracy of the reading, but that has a dangerous side effect: The base of the simulator is going to be at burning temps! Guessing again a 1w/mm2 heat load and that his brass die has a height of about 20mm, If he tests a cheasy heatsink that's only cappable of keeping the die surface at 60ºC, the base of the brass die is going to be at 227ºC! Things are going to be worst than that at the cartridges un the aluminium base.

Robotech: If I were you, I'll reconsider the use of brass, or at least lower a lot the height of the thin part.

[nicozeg]

Edit: Damn double post!

[Groth]

Well said, nicozeg. That's precisely what I how I was thinking.

For the calibration, I'd also throw in a test with the die surface air cooled. Readings from an IR thermometer could be compared to the expected value.

[bigben2k]

Originally posted by nicozeg
By choosing brass for the heat die the gradient have a big increase and improves accuracy of the reading, but that has a dangerous side effect: The base of the simulator is going to be at burning temps! Guessing again a 1w/mm2 heat load and that his brass die has a height of about 20mm, If he tests a cheasy heatsink that's only cappable of keeping the die surface at 60ºC, the base of the brass die is going to be at 227ºC! Things are going to be worst than that at the cartridges un the aluminium base.

That would make it pretty difficult to insulate, no?

[gone_fishin]

ok, you spend big money for .01C resolution and .05C accuracy then immediatly through your implementation you halve its accuracy which means there is a possibility you could be off by as much as .1C. Kinda makes paying for .01 resolution meaningless because you will never see it.


Is this the kind of equiptment we're talking about? $400 plus the sensors and readouts X 2.
http://www.microdaq.com/mt/temperature/rtdtemp101.php

[RoboTech]

Originally posted by nicozeg
Robotech: If I were you, I'll reconsider the use of brass, or at least lower a lot the height of the thin part.

Yes, you may be right... My initial concern with using copper was that the top surface of the die would be too soft and would quickly become deformed with repeated mountings. I do plan to trim the height down - I intentionally left it long so I would have material to work with when I get it mounted and lap the die down flat and parallel with the socket surface.

Maybe I will re-do it in copper... short, truncated pyramid out of copper with a thin (0.12") brass top plate??? Or maybe I should just forget the brass bit all together??? Copper seems to be working for a few others.

Thanks for the heads up...

[nicozeg]

If copper softness is your concern, you can use some type of dural for it's hardness and better thermal conduction than brass.

[RoboTech]

Originally posted by bigben2k
So does anyone know what the space shuttle uses for thermal insulation, and if there's anything similar we can use?

Yes, the ceramic thermal tiles on the space shuttle are made from very pure amorphous silica...

http://science.ksc.nasa.gov/shuttle/....html#sts-hrsi

For a more practicle (and affordable) soultion for insulating thermal dies take a look at Cotronics; specifically their flexible ceramic thermal insulation products.

http://www.cotronics.com/vo/cotr/fc_1.htm

Ceramic blankets, board, moldable sheets and papers (refractory fibers - alumina oxide/silica oxide).

[joemac]

Originally posted by sevisehda

Groth: That is one brilliant idea. It may be too perfect though, at times its difficult to make the math fit the real world and this may be one of those times. Most likely it would would but it would need to be calibrated quite accurately

I agree

Originally posted by sevisehda
Then calculatte the amperage by V=IR.

Don’t you mean I = V/R

[joemac]

OK check this link out :

Water Block Testing Alliance