Error convergence of mesh optimization

[DPeixoto]

Hi everyone,

Although I just signed up, I've been reading posts on this forum for a while now and I've found them quite helpful! Hopefully you guys can help me just as much. I am performing a thermal analysis on a set of cork-core sandwich panels with aluminum facesheets in ABAQUS. The sandwich panels, which are laterally isolated, are heated on one side/facesheet and the objective is to know the temperature on the other side.

The problem here is the mesh refinement. I performed various analysis with different meshes and ploted the value of the facesheet temperature vs. number of elements and as you would expect the value converges. This however is not enough to justify why I chose a specific number of elements for my mesh.

Is there any statistical method on Matlab that works with a desired value of the error of the temperature vs. number of elements function?

Thank you in advance,

Best Regards,

DPeixoto

 

[GregLocock]

If you do an error analysis on your measured parameters you will come up with an error band. Then choose a mesh density that gives an answer relative to the converged answer that is within that error band.

No you can't predict the error from the mesh density for a general geometry, that is effectively doing the analysis before you've done the analysis.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[DPeixoto]

Thank you for the advice Greg! I'm not sure how to do the error analysis though. Do I have to perform the error analysis on MATLAB?

I already ran a set of analysis and came up with a temperature vs. no. of elements graphic!

Regards,

DPeixoto

 

[GregLocock]

You have assumed certain values for geometry material properties and boundary conditions. Each of these has an error associated with it.

http://reference.wolfram.com/applications/eda/ExperimentalErrorsAndErrorAnalysis.html

might be a good place to start

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[DPeixoto]

Thanks for the link!

I was recommended using the richardson extrapolation to get the optimal number of elements for the mesh. Do you happen to know how somehting about this method?

Regards,

DPeixoto

 

[GregLocock]

No

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[GregLocock]

Well OK, that looks dead easy

You need to write a function that runs your FEA model, having scaled the mesh density by a factor Smesh, and returns one output value and then use that in the easily found matlab code for a richardson loop.

https://ece.uwaterloo.ca/~dwharder/NumericalAnalysis/12Differentiation/richardson/matlab.html

The richardson bit is not the hard bit.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[DPeixoto]

I used the matlab function that I got from the wikipedia page, but unfortunately it didn't work. The function that runs ABAQUS however, how do I know which function it is?

Thanks again for your input and for helping me out, Greg!

 

[GregLocock]

not being an abaqus user you'll have to work that out, in my case I call my solver from a dos bat file using this line

for cpu=1:num_cpus
if run_adams==1
dos(['runadams' num2str(cpu) '.bat &'])
end
end

and runadams1.bat looks like this

call achassis -custadams C:/adams/Cust_ADAMS/R4/Personal_Py/apresolver_R4_x64.dll glts1.acf
!call needed otherwise dos box doesn't close
exit






Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[DPeixoto]

Hey Greg,

How do I get this .bat file? It seems to me that the abaqus job file have either .inp or .odb extentions.

Also, I found a code that supposedly works with abaqus. How do I integrate the richardson extrapolation code with the abaqus code? (By making my function f(x) being the abaqus code below perhaps?). The code is as follows:

% Running regular jobs from within Matlab
% Note: These commands worked on a Windows Vista 64-bit PC.


% Running simple INPs - one after the other (2 here) - in batch mode:

clear all; clc;

dos('cd "DIR1" & "DIR\temp1.bat" && "C:\WINDOWS\system32\cmd.exe" /k exit')

dos('cd "DIR2" & "DIR\temp2.bat" && "C:\WINDOWS\system32\cmd.exe" /k exit')

% ..

% ..

% DIR1, DIR2 .. are the locations of the directories in which your INPs, temp1.bat,
% temp2.bat .. files respectively are placed

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

Best Regards,

DPeixoto

 

[DPeixoto]

Ok, I think I got the .bat file part. The .bat file is supposed to be a file with a command that runs ABAQUS and the desired job.

The question now is, as you already pointed out, how do I extract the desired output values (Temperature vs. Number of elements)?

Is it possible to run manually the same job a certain amount of times with a different number of elements each time and put the resulting temperatures in a table a use that table in the richardson extrapolation?

Best Regards,

DPeixoto

 

[GregLocock]

Yes for an optimisation to work you need a single score figure. So read your output into matlab using textscan or whatever, and analyse the results

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?