top of page
If using this toolbox for research or industrial purposes, please cite:
Advances in Engineering Software. Vol 105. March 2017. Pages 9-16. (2017)
Abaqus2Matlab v.3.0
A new way to post-process FEA
bottom of page
Thank you so much George. Thats a very good information. If you don't mind can you please look at this matlab post and comment on whether the way I am giving the inputs and targets in the cell format is correct. Because my final curves are not matching. All the parameters are going into a local minima and are not updating after some time steps.
https://in.mathworks.com/matlabcentral/answers/468697-how-to-apply-neural-networks-for-multiple-experimental-data?s_tid=prof_contriblnk
Dear Bharat,
Apologies for the delayed answer but I am out of office these days because I am attending a conference.
You can use the same objective function that is used in the example by adding the contribution from other experimental curves, i.e. by adding terms of the following form:
...+abs( ( a(n+1)- a(n) ) / a(n) )+... etc.
where a(n) is the value of the fitting coefficient "a" in iteration n, and a(n+1) is the value of the fitting coefficient "a" at iteration n+1. You can add as many such terms as you want, depending on the number of the experimental curves that you want to fit. You can also place some weighting coefficients is some of these terms, if you want to emphasize in specific experimental curves.
In general there is not any uniquely "correct" objective function for these cases; the objective function is usually problem-dependent. You should use an objective function formulated in a way that leads to the correct result with the fastest convergence rate possible. Trial and error will show you which objective function is best suited to your needs.
Best,
George
Hi George, Any suggestion for objective function??
Hi George, Thanks a lot for creating this excellent tool. Infact I went through the tutorials in Abaqus2Matlab and I am really happy to see the curve fitting in cleavage model. You fitted single curve in this model. I want to modify the same script for my problem to fit multiple experimental curves simultaneously by keeping the parameters constant for all the curves. So, I was wondering whether I can use the same objective function by adding the contribution from other curves. I was confused about how to give the multiple targets (I mean experimental data for different curves) for the same input. Can you give me any suggestion regarding these issues. Thanks again.
Regards,
Bharat
Hello,
Thank you for your interest in Abaqus2Matlab.
Fitting experimental curves with Artificial Neural Networks (ANN) includes:
(1) Obtaining the training data for the ANN by performing a number of Abaqus analyses and transferring their results in Matlab using Abaqus2Matlab
(2) Training the ANN in Matlab based on the training data above.
You can check the following links for more information about coupling Abaqus and Matlab:
http://abaqus2matlab.wixsite.com/abaqus2matlab/forum/main/comment/59a97f7c416d360010569486
http://abaqus2matlab.wixsite.com/abaqus2matlab/forum/main/comment/59d15bbb16f6ac00ba2295f8
http://abaqus2matlab.wixsite.com/abaqus2matlab/forum/main/comment/58cb995c9a2f44020905ef7b
To automatically obtain the training data in Matlab, you should create a MATLAB function which creates the Abaqus input file which is used in the Abaqus analysis. After this, you can easily use this function and other suitable Abaqus2Matlab functions within a for loop, which will give you one pair of training data per repetition.
I hope that the above will help you. Please ask any question and share any comments or suggestions that you may have.
Best regards,
George