How it works.

[gmat]

Or how i'm planning to do it.
The program will do a series of passes, each solving a particular problem.

1 - solving flow: the Pressure/Flow data of each component will be matched and the program will output a resultant flow. From now on we'll assume the flow is fixed, ie not dependant on external factors or temperature.

2 - Air flow - for the rad. Need Flow / Pressure data (of the chosen fan)

3 - Heat dissipation for the rad, air side. Using air flow / air dissipation data and ambient temp.

4 - Heat dissipation for the rad, water side. using water flow / water dissipation data

4.5 - correlate the heat impedance for the rad.

5 - Heat impedance in the WB. Using Flow / Heat resistance.

6 - Water temperature: this will take into account the heat impedances of elements vs the heat source and vs the heat dissipation of the rad.
(do we need to consider the sequential aspect here ?)

7 - heat source temperature.

----
At every stage no formula is involved. I know some ppl here want to write formulas for elements, but writing a parser for math formulas is a diffcult task, and not a goal of this project.
The program will use 100% interpolated data, actually it will build polynoms upon a set of data points provided in XML files of elements.
Each solving stage will use Newton's method.
---
Did i forget anything ? (or made any mistake ?)

[Since87]

Sounds good, but a few things:

3&4) Will we be able to make that 'split'? Or will we need a 'two dimensional interpolation'? Rt = f(Qair, Qwater) ?

6) Handling the 'sequential aspect' would be ideal. The dT, inlet to outlet, is a large enough fraction of the dT, water to air, that it will be significant. Also, considering the constant pump->rad->block vs pump->block->rad discussions.

I think it can wait for a refinement stage though. The simulator can still provide useful insight, with water temperature considered to be equal throughout the loop.

[BillA]

Originally posted by Since87
. . . . The dT, inlet to outlet, is a large enough fraction of the dT, water to air, that it will be significant. . . . .

huh ?
if you have a +0.2°C rise in the wb coolant temp, that (plus perhaps 0.05°C from the pump) is exactly the drop you will see in the rad

"significant" ?
did I misunderstand something ?

all we are doing with this 'calculator' is solving for the coolant equilibrium temp,
then taking the wb's "C/W" and back calculating a CPU temp

[Since87]

Well, I didn't say, "very significant".

The error resulting from neglecting dT's within the loop, would be well below the measurement resolution of most people.

[deeppow]

Originally posted by Since87
The error resulting from neglecting dT's within the loop, would be well below the measurement resolution of most people.

What? If the temperature change is unimportant then there is neglectable energy transport. The fluid side must be far oversized and you're controlled by the air temp for the radiator which will be your minimual fluid temp (same as air cooling). Why bother to worry bout the water system design at all?

[Alchemy]

Originally posted by deeppow
What? If the temperature change is unimportant then there is neglectable energy transport. The fluid side must be far oversized and you're controlled by the air temp for the radiator which will be your minimual fluid temp (same as air cooling). Why bother to worry bout the water system design at all?

I think you misunderstand. The change in temperature of the water between any two points in a water cooling system is very small because of the high heat capacity of water.

The difference in temperature between the air and the water or the water and the CPU core is, of course, very significant.

Alchemy

[deeppow]

Originally posted by Alchemy
The change in temperature of the water between any two points in a water cooling system is very small because of the high heat capacity of water.

The difference in temperature between the air and the water or the water and the CPU core is, of course, very significant.

Actually I don't think I misunderstood.

The water can't be cooler than the air unless we're using phase change or refrigeration of some type, e.g. evaporative cooling of the air going into the rad. Thus the temp of the water at the entrance of the water block will be at room temp at best unless other mechanisms are used.

This does raise an interesting thought. This line of thought says one of the problems in testing required of BillA, or whomever, is the accuracy for temperature measurements in this situation. Measurement of the energy transferred from, and to, the fluid is really not possible via the fluid temp, or extremely difficult at best. What about a simulant fluid with less heat capacity so temperature can be better used? This is done all the time in many experimental applications and must have been thought of here.

[Alchemy]

Originally posted by deeppow
Actually I don't think I misunderstood.

The water can't be cooler than the air unless we're using phase change or refrigeration of some type, e.g. evaporative cooling of the air going into the rad.

True.

Thus the temp of the water at the entrance of the water block will be at room temp at best unless other mechanisms are used.

Yes, and will probably be somewhere around 25-35 C.

This does raise an interesting thought. This line of thought says one of the problems in testing required of BillA, or whomever, is the accuracy for temperature measurements in this situation.

Yes. Temperature measurements on the range of 0.1 C are going to be pressing the limit for accuracy for the fairly inexpensive type of temperature probe (thermocouple) It is quite likely a better thermal sensor like an RTD will only serve to pick up more noise and random fluxuations in temperature.

Measurement of the energy transferred from, and to, the fluid is really not possible via the fluid temp, or extremely difficult at best. What about a simulant fluid with less heat capacity so temperature can be better used? This is done all the time in many experimental applications and must have been thought of here.

Unless you know of a fluid with the exact same density, heat capacity, viscosity, thermal condution coefficient, etc., we'd need to make a host of correlations to compare the data between the two and introduce a lot of unnecessary inaccuracy and work.

Alchemy

[BillA]

Originally posted by deeppow
. . . . and must have been thought of here.

lol – not on this shift
we can't find the lint in our belly buttons

Alchemy
hell will freeze over before an accuracy of 0.1°C is achieved with a TC by you, me, or any else
and the budget does not matter in the slightest, don't give a shit what is spent
- you can't get there from here
the 'best' TC is a type T, and with a 'best' indicator you will be hard pressed to demonstrate ±0.2°C

now this has been covered by pHaestus, by me, and even webmedic
a suggestion: investigate before you state
(you are quite knowledgeable about some things, others may believe you also when you're blowing smoke, /rant)

RTDs are a whole 'nother kettle of fish, for enough $ you can see whatever is desired
my present setup has 0.01°C resolution and ±0.03°C uncertainity
- note that this is quite dependant on the actual RTD and its calibration (I use class "A"s)
and no, noise is nothing that a 4-wire setup and proper shielding cannot cope with

re the heat transfer via the coolant temps:
I don't do much of that at all, for the obvious reasons
(but I do record such as it is a good backwards ck of the coolant inlet temp)
pHaestus kind of has to do so as the CPU's heat quantification is being attempted

and I quite agree that all the correlations necessary to 'validate' a baseline for another fluid would be a massive source of potential error

[deeppow]

Originally posted by Alchemy
Unless you know of a fluid with the exact same density, heat capacity, viscosity, thermal condution coefficient, etc.,

Alchemy, you're suppose to be the professor so don't BS me. That is why they use dimensionless numbers. You don't have to redevelop the correlations that are developed using them as long as certain scaling parameters are satisfied.

Simulants are a compromise but they work. In this case I assume we would like to use the same hardware and stay single-phase-turbulent flow? Other things (flow rate and power for example) can be adjusted.

[opfreak]

i'm new here, but am seeing the same problem everywhere.

i think the models in general people are creating are 2 complacted to be sloved without hardcore pc power.

but i think a much more simplistic approach can be taken to the problem(s)

depending on what you want to solve for. the system for water cooling is very simple, more complex then just air. but not as complicated as we think.

to supper simplify we can, neglect heat transfer from the pump (esspically inline) and any heat transfer threw the tubing.

what we get is basicaly a 2 radiator system.

heres how it goes.

water flow is marked by ==>

rad1====>rad2===>rad1

we have a closed loop system, that after its running is steady state.

now from steady state anaylis we need to know a few things

a) mass flow rate of water, and the themral porpties of water. (from any themral book)

b) mass flow rate of air, and air temp( temp should be considered constant throughout a given system

c) the thermal energy being put into rad2 from the cpu

since almost all the system specs are known there are a few things that need to be added into the caculations to get accurate results, and that is efficency.

we need efficeny numbers for the transfer of heat from cpu to water, and from water to air.

these numbers can be assumed, or estimated, to try to get a worst case seniaro, or can be eventually obitaned threw data we get from people using various blocks/flow rates etc. then doing a best curve fit. also i'm sure some of these numbers are known somewhere, esspically for the rad for water to air. since these are ussaully car heat cores, and data for them should exist.

water blocks seem like a different problem, since finding data on them would not be possible as they probably have not been tested as much as the rads.

the top amount of heat from cpu, in watt/hrs can probably be found from like amd, or intel.


so anyway, what in the end we can find out, is most likely both water temps on both sides, and from the get an idea of what the cpu temp's will be like. i'll work on equations later, but after there sloved should be plug and chug

[Alchemy]

opfreak, you'll learn a lot of you read the other threads in the Water Cooling Simulator Discussion topic, and maybe a few others on the boards. There's a lot of good info on basic heat transfer and fluid flow on this board, and a handful of people very skilled in this stuff.

If you have questions about how any of this works I'm sure you can get a dozen of us to help explain the basics.

Alchemy

[gmat]

opfreak, all this is taken into account, as i'm using a 'black box' model for each element.
The solver uses Newton method to solve the n-th order polynomial produced by the combination of all elements.
The program will accept other interpolation+solving methods of course, we're just starting with polynomials as they seem to fit quite well, and because it's easy.

-- quick status update --
I'm progressing slowly currently, not because it's out of hand, but mainly because of Real Life. It's only a home project and i need free time for that. Hopefully i'll get some vacation (2.5 weeks before the end of May) so i'll have spare time to code (i really hope so).
I'm also setting up a good base so people can contribute easily with modules, datasets, interpolation methods, solving methods, without an acute knowledge of Java or of the whole project.
I have already all libraries (thanx to Apache project), the code is on CVS, the source tree is ready, the module parser is done and i'm writing a few test cases currently. After that i'll write the solver.
The hardest part on this project will be the user interface, and most of all the creation of datasets (ie real heat transfer data)

[opfreak]

ahh like i said i'm new, i just read a few postings and didn't know where u guys where. don't mean to step on any toes.

i hope ur using a program like matlab to solve those nth order polynomials. otherwise it just seeems like a pain

[gmat]

about matlab: it's not the idea for this project. It will be a standalone app, and it's written in Java and XML. The goal is to allow everyone to use it and extend it.
Besides solving nth order polynomial aint rocket science. Newton's method sounds like a good idea and writing it in Java is easy enough.
Trust me there - the tough part is *not* the maths or formulas. It's the User Interface.

[opfreak]

ahh, ok, now i see wat u guys are doin, my bad

[LiquidRulez]

It'll be interesting to see the results of this project indeed.

[bigben2k]

Originally posted by gmat
Trust me there - the tough part is *not* the maths or formulas. It's the User Interface.

Oh man, do I hear you!

Are you going to make any effort into making it "idiot proof", or are you just going to prevent zero divisions?