Numerical Modeling of Water Blocks

[Aesik]

The results are all coming from finite difference code written into a spreadsheet (see the very first post for a synopsis of how this works). All of the governing equation and conditions have been hand coded by myself. This is where I thank my engineering professors for making me do all those seemingly useless projects before.

 

[Maximus Nickus]

One word,
Superb!


M_N

 

[vandersl]

Great work Aesik.

I have a couple of questions.

How are you modelling the fluid flow? Do you have a fixed geometry, and assume a certain heat removal rate at the WB/fluid junction for a given temperature delta and flow rate? Or do you actually model fluid flow in the channel (i.e. turbulence, water warming up as it flows through, etc)?

I am asking because I made a program like this a little while back, but it is somewhat simplified. It allows material specifications (density, thermal resistance, heat capacity) and geometry built up of nodes of any specified material. Where it really falls down is modelling fluid flows (natural or forced convection). Instead, I simply model air or water as having artificially large heat capacity (equivalent to an essentially infinite flow/removal rate), which seems to give me reasonable outcomes. I am still looking for a better model that doesn't involve too complex a calculation.

I'm not real strong on my modelling skills, so I don't establish initial and boundary conditions. I simply start with some initial conditions and iterate until I am happy with the outcome (i.e. steady state has been achieved). To meet stability criteria at very small node volumes I need to go to very small time increments (1ms or less), so you can imagine that it can take a while to run even a few seconds of simulation.

 

[Aesik]

The fluid flow is modeled by taking all the physical dimensions, constants and properties:

Channel dimensions (height, width, length)
Conductivity
Prandtl number
Density
Viscosity
Flow Rate
Input Power
Input Fluid Temperature
etc.

Using this, it calculates hydraulic diameter, velocity, mass flow rate, Reynolds number and fluid temperature delta. Then the Gnielinski correlation for turbulent flow is used to find the friction factor, Nusselt number and the convection coefficient. All of this is then fed into the elemental model of the channel. It sets up all the boundary conditions such as input power to the area directly over the cpu, fluid temperature along the path, and convective effects over the surface in contact with the fluid. Then I just set up the iteration and convergence criteria, enable circular references, hit F9 and watch the fun (actually, watching paint dry is generally more exciting).

Once the model has converged, it spits out average temperature of the area in contact with the cpu for however many baseplate thicknesses it's setup to run at the time and graphs it. I don't have it setup to automatically graph the temperature profile across the whole block because that takes too much computational time. If I want that I just cut and paste the values and use a surface plot.

Right now I only have the model setup to assume turbulent flow. I'm not sure how useful including laminar flow would be because most systems will have at least some degree of turbulence. I'll probably eventually add laminar cases for completeness. In the future I'll be adding a function that allows for geometry induce increases in Reynolds number, but I'm still studying up on how to model that correctly. That part is 49% engineering know-how, 51% black magic.

 

[dream caster]

Aesik:
I don't understand completely the kind of model you have, how many channels does it have? what are its phisical dimensions (width, depth, channel dimensions)? and wich heat paths within it are you considering?

BillA :
can you give us here or in another thread some summary about the perfomance changes with geometry you have found ?

 

[Aesik]

Dream Caster, I think that if you read the thread thoroughly, you will find all your answers and more.

 

[dream caster]

Aesik:
this i gathered from you:

...my a model ... Just a long, straight tube. Not to mention that there are no parasitic heat losses built in...

...This is for a block with channels .25" inches square and 10cm in total length. I varied the base plate thicknessfor both an aluminum and a copper block...

...Setting up the model is the easy part, getting the boundary conditions right is a whole other story and has caused more than a few mental breakdowns through the years...

what worries me is the "boundary conditions", we cannot say how useful to us is your model if we do not know them and from what i quoted I do not know them. I think the biggest problem when one establishes a model lies in the basic assumptions: knowing all those that have been made and if they are useful descritions of reality or not.

 

[Aesik]

Sorry dream, this wasn't the original thread so you've probably missed some of the conversation. The missing conditions you are looking for are input power of 100 Watts and input water temperature of 25 C. There will be lots of data coming soon with all sorts of conditions and such. Right now the model is just evolving; it's no where near finished yet.

 

[dream caster]

I think you guys will be really interested in this procooling thread, Water Blocks roundup of my own about some similar but not equal blocks built by the same guy and tested by him. Also near the end there are some model results about base plate thickness that are similar to those from Aesik's model.

 

[Aesik]

Just a little update on this project for those interested:

I've worked about 22 days straight (that would be real work, my job =P) and have been pretty much burned out on doing much of anything lately. However, this little project keeps gnawing at the back of my head so I decided to take the next step. I've abandoned using the spreadsheet because it requires that every geometry be setup and cannot be dyanimcally changed. I'm now working on a Fortran program that will allow changing of all the variables on the fly without having to worry about hard coding any geometric features. The other great benefit is that it now takes seconds/minutes to run a full blown simulation whereas before it was getting upwards of hours to do so. Just goes to show what bloat-ware MSExcel really is!

There is still alot of work to do, but soon I should be able to post a copy of the program for your pleasure (if you're masocistic and need to kill some time). The first version will still be the simple model allowing variations of flow rate, input power, length, height and width of channel, etc. I'm working now on trying to make it more user friendly and also to be able to loop multiple simulations simultaneously to save time. It won't be anything pretty (yet) but will be a program that can be run from a command prompt. It will spit out a few numbers to the screen, and print all of the resultant nodal temperatures to a text file that can be imported into another program for graphing. One of these days I'll get off my duff and figure out how to integrate Fortran code into windows apps to make it slicker.

If you have any ideas to add to my wishlist (see previous posts for what I will be adding) please let me know and I'll see what I can reasonably do.

 

[dream caster]

I'm very interested in seeing all that.

 

[nihili]

Just a little update on this project for those interested:

If you have any ideas to add to my wishlist (see previous posts for what I will be adding) please let me know and I'll see what I can reasonably do.

Well, I've got plans for a variable flow water block milling around in my head. And I'd like to base it on a graduated thickness baseplate. So if you could add that in and give me an educated guess on the best angle for the gradation, that would be tres cool.

nihili

 

[BillA]

I'm going to focus a bit on where this could wind up

every (worthwhile) wb today has its inlet and outlet normal to the bp

the inlet options are but 2: just off center (wrt the die),
and at one end of a channel whose mid-point passes over the die area

internal configurations range from nothing (the 462), to a 'ducted' flow (the Maze3), to a labyrinth with internal turbulators (the new Dtek)

bp finishes range from smooth, to sandblasted, to 'grooved', to dimpled

so why would a CFD modeling program not reflect these same parameters
AND running an analysis of an existing wb enables us to validate the mathematical model
(without which we are just playing make-believe)

be cool

LATE EDIT:
a suggestion as to how to actually proceed - make a CFD model of the most simple wb available
- I would suggest a 462
the bp can be instrumented with embedded TCs and the predicted temp distribution compared with the actual
then the velocity under the inlet can be changed
then the flow rate, etc

 

[Spartacus51]

First of all I must say great work everyone, very valuable thread, and thanks for all your time and efforts.

Secondly, with all the great data in this thread can we get a post on conclusions? Obviously most people here are capable of reading a line graph. My question is regarding real world application. While my pump is rated at 300 GPH I really doubt I'm getting that... perhaps only 100 GPH. Looking at the graphs I see that with my copper WB I should have a 10 mm base... well the whole thing is only 12 mm. Do I need to get a bigger chunk of copper and make a new one? (In lieu of getting a better pump) Or will the structure of a waterblock allow for a thinner base?

Again, thanks and keep up the good work.

(Oh, to have about 2 more years of calculus and a thermodynamics class)

 

[BillA]

this is all 'what if'
we are a long long way from a design guide

be cool

 

[gone_fishin]

What about graphing a variable base plate thickness for copper. I have a cone design over the core type block I plan on attempting. Instead of making multiple blocks I would like to guestimate the optimal diameter at a set height of the cone at 1/2" for a given wattage.

 

[Spartacus51]

this is all 'what if'
we are a long long way from a design guide

be cool

hmmm, looks like I raid the shop tomorrow and grab a bunch of 1 inch copper and start making blocks... we'll see if I get results similar to the "what if"s then.

 

[BillA]

the relative relationships should be roughly ok,
just no absolute accuracy in the values

be cool

 

[BBigJ]

Be careful what you guys ask for. If there is any problem more difficult than heat transfer it is turbulent fluid dynamics. Some of the very simple blocks (like the swiftec) may be the most difficult to model because the flow path is unsolvable (although if you can find a good way to solve it, go ahead and pack your bags for Stockholm to pick up your nobel.)

Aesik, personally I'd like to see an option to adjust the flow direction.

 

[BillA]

BBigJ

perhaps exactly the question I've spent the last 10 days asking - w/o any answer

is it not possible to model the impingment of a jet at right angles to a surface ?

you know anything about this ?

my only datum is from a PhD candidate (laminar flow dissertation) who says "virtually impossible"
(and he has no means to get authorization for the prodigious amount of computer time needed)

why ?
anyone with knowledge of such ??

be cool