Temperature measurement
 

Following discussions on the heat die here, as well as other bench requirements, here is where we discuss temperature measurements.



A recent related question came up (here) about the type T versus type K (last page).

pHaestus: there's an outstanding question to you, about type T versus type K thermal probes. Do you want to answer it?

http://www.microlink.co.uk/microlink/tctable.html

I assume 0-200C is of relevance here. It's also easy to make type Ts with very thin wires if that's your thing (see JoeC). I like my dual linear thermistors more than thermocouples, but the Type Ts are handy for inserting into wb baseplates and die simulators.

[edit: rambling removed]


The temperature measurements is a really critical component of a test bench, so I'd like to tackle that question, and get to some kind of idea as to what level of accuracy we're all going to shoot for here. Everything else falls back on this.

Well the key thing is to have some way to periodically check your probes and calibrate them. This means that you need a single really accurate temperature monitoring device for cross calibration, a device specifically designed to standardize/recalibrate, or you need to budget for getting them professionally calibrated periodically. I honestly don't have a ton of experience with thermocouples but have been told before that linearity is a concern. It isn't like a thermistor where there's usually just an offset when they are calibrated, right? They work through Seeback effect and the equation is much more complicated.

On a budget, I suspect that something like a Fluke 2190 (get the multiplexer and you can have LOTS of probes) and type Ts is about as good as you can do.

For price/performance on temp monitoring, do not forget that being able to dump the readings to a PC is worth some money. I have analog temp readers now and it is a hassle.

Oh and as far as temperature measurement discussion goes:

You will need to measure (at a minimum):

water inlet
water outlet
waterblock baseplate
die

I also measure ambient air and radiator outlet temperature (just because I have the capacity to do so).

What accuracy is truly needed? I hear words like analytical vs comparative batted around in e-mail exchanges. Are you distinguishing the two based purely upon the accuracy of temperature measurement?

Although I can see the desireability of knowing all these temperatures and more, I'm not sure I see the "need" to measure more than the die and the water inlet temperature.

Water outlet temperature can tell you the amount of heat the waterblock is dissipating, but assuming a 'perfect' die simulator, I'm not sure it is a necessity. (Obviously this test should be done to measure the quality of the die simulator.)

If it turns out that the die simulator 'leaks' too much heat, it may be desirable to make the delta T between inlet and outlet water the measure of heat transfer. This requires extremely good accuracy and resolution of the temp sensors though. (Not saying that extremely good accuracy is not needed anyway. Just that doing this would make the need greater.) It also requires extremely accurate flowrate measurement.

Alternatively, the water outlet temperature could be ignored (during actual waterblock testing) and 'leakage' of the die simulator characterized. This would require an ambient measurement though. (Since 'leakage' will vary with the deltaT between ambient and die temp.)

I don't see the value of, or a way to practically implement, measurement of the WB baseplate temperature. Few waterblocks of major interest these days are going to allow for placement of a temp sensor in a useful location.

Since87:

You are right; the baseplate and water outlet temps are mostly for the tester's internal checking and peace of mind and they aren't carried forward in any calculations. But here is how I am using them:

I find the baseplate temp useful as I compare to Bill's previous data (I am using a block that he also tested so the hole is in an identical spot). With my CPU I just don't know the power applied and so I am using the delta t die to water, delta t die to baseplate, and delta t water out to in all together to see if I can things to correlate. I guess this isn't explicitly needed when everyone is using die simulators, but then again if every unit is insulated differently...I also use the baseplate temp to track my reproducibility at waterblock mounting/paste application. YMMV I guess on the usefulness of it.

I look at the water inlet and outlet as important from a heat balance standpoint. If your die simulator is "leaking" then you can pick that up with these measurements pretty easily. If you don't do a check like that then you are going to have some pretty widely varying C/W values among different benches even with good thermal monitoring. This is probably why heatsink testing from commercial sources reports such widely varying C/W values.

Yes, that makes sense. I think one thing many people who are thinking about testing water blocks don't consider, is that you need more than the equipment which composes the test bench. You also need additional equipment to verify the operation of the testbench.

Ok, to recap:

-Need to measure:
1-the heat die temp
2-the block water inlet temp

-need a highly accurate probe, for periodic calibration.

-Alternate temp measurements:
1-block outlet (highly accurate, to measure the heat flux)
2-ambient air
3-water block baseplate (where available)
4-radiator outlet (water)


[edit: rambling removed]

Do they still make these? Anyway I was going to buy one of these http://www.tequipment.net/FlukeThermo54.html The 54-2 series with data logging and dual probes. but it is only dual probes and it costs $300.

If you look here at the price page: http://www.tequipment.net/FlukeThermo54MO.html
and scroll down you will notice the probes are not exactly cheap either.

Probably not the lest expensive place to get this stuff though.

I was just looking at the same one.

The 50 series can return a value within +/- 0.3 deg C, regardless of the thermocouple used. The T type probe though, would have an absolute error of +/- 0.5 deg C, where a type K would be +/- 1.5 deg C.

[edit: rambling removed]

The different models have various extra features, like logging, dual input (and differential output), etc...

Yeah, my second plan was to buy one of the cheaper ($150ish) Fluke 50-2 series and just use it to calibrate the sound card mod thing OR I was going to get one of those maxim kits pH is using.

My head is killing me today. Damn cold front and rain! Bah, back to work....

Ben, if you can't answer this question yourself, what hope is there, that you are going to be able to cover all the bases in setting up an accurate test bench?

Which is of course, absolutely useless when trying to measure the heat dissipation of the block by measuring the delta T between inlet and outlet water temperature.

And if you don't measure the die simulator's secondary losses, then all of your measurements will be suspect.

[edit: rambling removed]

since87: GREAT link. Informative +1. The comments about sources of error are most useful for tc users who are buying parts off ebay.

Ben,

Slow down and think a little before you post. I realize that rambling is one way to work things through, but I generally keep it to myself rather than putting it to paper, so to speak.

To your original question, yes, increasing the ambient will increase the amount of heat leaving via a secondary pathway. That should be self-evident as the delta-T from the die to air drives secondary losses. Assuming the air remains constant and you raise your fluid ambient, more heat will transfer to air. Unfortunately, it is not so simple as assuming a constant convection coefficient to the ambient air as you are off in the realm of "natural" or "free" convection where a significant portion of the air velocity is a function of the density change as it warms. You can pot-shot it and take copious measurements to back up your guestimate.

A rather fundamental thing is that the repeatability of pretty much any online temperature sensor will be worse than the delta-T of fluid through the block. This has been said by several here, but it isn't apparent that the meaning has truly sunk in yet.

As for measuring secondary losses, one way is to install a well-insulated block with no flow. Then you modulate power to the die simulator and measure temperature of the die. You get a feel for how much power is dissipated vs die temperature when (almost no) power is going into the block. The obvious goal is to insulate the simulator such that you can scarcely turn the power down low enough to avoid a high die temperature.

[edit: rambling removed]

Hey Ben,

IMHO thermocouples (whatever type) are good for applications that don't require high accuracy. Even with a quality TC and calibrated instrument you won't be able to do much better than about +/-0.5 deg C. TCs big advantages are they are inexpensive and widely used throughout industry (available in all shapes and sizes). If you need better precision and accuracy then 4-wire platinum RTDs are the next step up. They are more expensive but in a best case situation can provide +/- 0.01 deg C accuracy.

So I recommend looking at RTDs for the die temp and water inlet. Don't even bother with TCs. Even with good RTDs installed on the inlet and outlet of the block, I doubt we will be able to measure the Delta T across the block accurately enough enough to define the actual heatflow. To define Q based on mass flow rate and Delta T would require even more accurate (0.001?) temp readings and very accurate (0.1?) flow rates.

In regards to calculating the thermal die secondary losses - my plan is to...

1) insulate the block quite well (thermal stand-offs, air gaps and ceramic insulation) to help minimize secondary losses as much as possible.

2) Measure the secondary losses by firing up the die and applying just enough power to bring it up to a typical steady-state operating temperature (i.e. 40 C) at a known and controlled amb air temp (with the die surface insulated). Once the die equillabrates at temp, virtually all of the power going in (E x I) is going out via secondary heat paths (since there is no WB sucking heat out the top). Since I'm not far enough along to have actually done this I don't know how well it will work.

Well, so much for lunch... :)

Going over RTDs now. That Omega site has some info.

I should also point out that Bill used an RTD in his original heat die.

You realize you are doing this in exactly the wrong manner, right? I would suggest:

1) Come up with goals for the testing
2) Run some numbers with different levels of accuracy in the temperature and flow rate and power measurement.
3) Price out gear that will get you where you want to be error-wise. This will let you meet your goals with the minimum outlay of cash.

I get the impression that your goal in #1 is "to get the exact correct answer", and that's going to result in a $ value for 3 that's near infinity.

I also get the impression that there is a bit of snake oil being sold in this whole WBTA thing. Quality testing just can't be done without a big investment in both time and money.

An allegory for the interested: I recently made some arsenic samples of known chemical composition in my lab. I started with a 1000mg/L standard that I diluted to 0.1 mg/L using an analytical balance accurate to 0.1mg and with density of water adjusted using room temp. I sent these to 4 professional laboratories in my region and asked for As analysis. The point? (1) none of the labs reported 0.1 mg/L (but one was "close enough" with 0.12ppm) (2)%RSDs were 1-2% max (3) there was a range from 0.02 to 0.4mg/l for the concentrations.

So what happened there? Each lab used probably a $100,000 instrument with a professionally trained technician. Each lab had good reproducibility of measurement, and none of the labs were correct.

Rather than spend thousands of posts on talking about the minor details, I would suggest starting with error and uncertainty and how to propagate error forward. That would be my first standard if I were a WBTA; all points must have error bars. Equipment needed would all follow naturally from that standard.

[edit: rambling removed]

[edit: rambling removed]

I don't know....

What I am not sure about is why we need to make such an analytical approch to this. Do we really have to know what the block does through out various flow rates? Do we really need to know the Base plate temp? Do we really need to try and measure inlet/outlet temps?

I just want to tell of block A is better than block B on my test bench with in a resonable amount of error. I was planing to sort blocks by groups. Group A would fit into high performance, Group B in middle performace, Group C low performance, and Group D not worth the metal used to make it.

Do I really need .01 accuracy to do this? Seems to me this is all the readers want. I don't think they want to hear all the flow rates, pressure drops, blah blah blah... They just want to know what level the block is in. If it is worth their money or not and how wrong they might go if they buy something else. Remember most consumers don't know jack about what they are buying, they just want to be told how good it is without a lot of technical mumbo jumbo they not only can't understand but do not want to understand, and quite frankly shouldn't have to understand. They got better things to do as they should.

I don't know.....

Jaydee:

Normally I'd have some choice words for the importance of statistical significance and the central role of error bars in any honest testing. Luckily for you though there's now a waterblock testing alliance! And undoubtedly Ben will drop in soon to tell you that "it's ok there's nothing to worry about! There's plenty of room for fluff reviews and "good enough is good enough".

Hooray!

Truth is you have to be mentally off kilter to spend thousands of dollars to test $40 waterblocks. I've cobbled together an almost acceptable test rig for a few hundred, and it's a pain to use. But I enjoy going downstairs and fiddling with it and making my little at home heat transfer discoveries. It will be really satisfying to have it all come together and post a new waterbock comparison article here on Procooling. Would you honestly be satisfied if that article DIDNT include a pressure drop vs flow rate chart? Or a C/W vs flow rate comparison for all blocks? The readers of this site are sophisticated enough to deal with that I think. Now the statistics behind whether a line running through points on a graph is meaningful or complete bullshit may be a bit too much for the average reader of this site. That doesn't make it any less important a question though...

'High' not "low", and what the dT can get down to depends on how restictive the block is.

I have already heard them and somewhat understand them, but still have yet to see any real evidence it is usefull to the average Joe.
Your refering to procooling.com and I am not. I "guess" I would expect those things to be in the review here as maybe 2-5% (that percentage unfortuantly doesn't seem to be rising, but falling) of the people on this site might be able to use those numbers to their advantage (if they actually do is another subject) and especially being we are more anal about things here. But I am not talking just about procooling.com. I am talking about my own site and the hundreds of other tech sites that do not have readers and never will have readers to the same level as a few here are. I see consistantly on many sites where readers get pissed off at how anal some are about testing. They don't want all this tech lingo and graphs. They just want to be told this is better than that and have a temp number from your test to show it. I feel this will never change as there is no other product in the world that has accomplished this even the one's with standards tagged on them like "UL" and "SAE" ect...


Maybe I am just not sure what "good enough" is anymore. If I can say with some reasonable certainty that block A is better than Block B on my system then I think I can live with that. I really don't see (yet) how the C/W value on my test system is going to be usefull for anyone being it will be impossible to replicate that C/W value on any other system (especially now that you say proffesional labs can't even do it with $100,000 equipment!). So I don't really grasp why it is important. Seems to me a temp number would be just as usefull and more easily understood to the average reader. Hell most people don't understand a lower C/W is better than a higher one!

But you say you have only got a few hundred into your system and you feel your capable of doing these measurments? If that is the case I will have to re-read your articles again and try and emulate your setup as I can handle a few hundred $'s! Ben is talking about what $9,000 just for temp monitoring? All I can say is **** that, I got better things to piss money away on strip clubs included! :D

I have been pondering what is good enough for a bout a year now when trying to put together my test bench, but I have yet to decide where I need to stop thinking and when to start doing. And if I start doing will it be good enough. :D

Bah, back to work....

This is exactly why I've been struggling with this issue of comparative, versus analytical testing.

In comparative testing, your only goal is establish which block is better, and its your total margins of error that's going to define your ability to do that. Otherwise, you just have to make sure that your testing conditions are similar, and as pHaestus pointed out, your results are going to fall within a range, which is best expressed using error bars.

In plain words, you'll only be able to tell that one block is better than the other if these error bars don't overlap. The more accurate you test, the shorter the error bar is going to be.

[edit: rambling removed]


Since87: thanks for the correction (fixed) (I hope that the meaning was still understood!) Nice catch!

That's right, you don't know. You will never get any grasp of performance unless it is done at various flowrates. One point on a graph seems to be all you need. Without resolution and accuracy your one plotted point now becomes a big blob on the chart, get the picture? So my question to you is, why bother testing at all if you are going to do it that way? How could you tell a customer what is better if you don't have a way of knowing? Just tell them anything I guess and back it up with a half assed tinkertoy test setup?

Sorry to come off harsh but you seem to be trying to cater to the masses, people are already doing that without any testing at all.