Why not MATLAB for FEA? 5

Hi all,
Instead of spending heavily on the Finite Element packages is it not possible to write code in MATLAB for the finite element analysis? Except those pretty picrues we can get good results I guess. I am just a beginner in FEA. I wanted to know whether one can write code in MATLAB as FEA has to basically to deal with matrices and MATLAB is a good tool for matrix manipulations. All this assuming one has strong fundamentals in FEA.
Comments awaited from experts.

simulator

 

[zuardy]

Hi fea_pandit,
To your question : "Instead of spending heavily on the Finite Element packages is it not possible to write code in MATLAB for the finite element analysis?"
Did you know, since when is MATLAB so powerful like now? To my know, it is still not long time ago. But the finite element programm written in FORTRAN (NASTRAN & ANSYS) did exist for more then 20 years at least.
Surely, it is not a great problem to write FEM-Program in MATLAB, even it is easier, then all Matrix functions are there. Some Universities offer their FEM-Programm written in MATLAB. Or some professional Software firms like :

http://www.femlab.com

has written MATLAB even for multiphysik problems. But don't forget, like all other engineering tools, the engineer prefer one tools that he knows well to newly developed tools (strange). I think, this is also the same case in FEM-Program.
Yet another thing : MATLAB is more slowlier compare to other programming language, isn't it?

cheers

 

[PriamEngineering]

For anything other than simple beam models this will be less efficient than traditional FEA. For non linear it will be very difficult as you'll have to decide on a convergence model as well as everything else. But I admire your inquisitiveness. I stopped asking these types of questions
years ago !

 

[mloew]

I can recommend the following book:
The Finite Element Method using MATLAB by Y.W. Kwon and H.Bang (2000)
Amazon:

http://www.amazon.com/exec/obidos/ASIN/0849300967/integratedsounds

B&N:

http://barnesandnoble.bfast.com/booklink/click?sourceid=320417&ISBN=0849300967

I used the codes in the book to solve some problems. The true power of using MATLAB would be to customize the code to perform some optimization routines and other specialized programming tasks within the MATLAB environment.

I no longer have the book or access to the code. If anyone has any code available, I would love to have access to it. I am trying to solve some simple problems, again, and this is the direction I would like to go in.

Best regards,

Matthew Ian Loew
m.loew@ssss.com

 

[brad]

One could certainly use MATLAB (or even hand-code in FORTRAN or other languages).

However, which beam (or shell or continuum) formulation are you going to use? If I am doing an analysis of a simple linear truss structure, I can certainly code that myself and trust the results. If I am doing an analysis of a complex structure modeled with multiple element types and with various kinematic assumptions, it may take me years to properly formulate the elements to describe the problem. By then, I'll have saved the cost of the commercial package, but spent more money in my own time than I've saved. I also will have to maintain that package, and as soon as a new problem comes along I will have to benchmark the package to assure that my assumptions initially coded are still appropriate for the new analysis.

In most companies, this is trade-off which drives them to use the commercial packages.
Brad

 

[abhio]

Can you really imagine running a 30000 element model in MATLAB? it would take forever! FEA programs are the only way to go for real world work.
Cheers
Abhi Abhijeet Oundhakar
Design Engineer
STUP Consultants Ltd.
Bombay, India

 

[zuardy]

not at all, if you use PVM (parallel Virtual Machine) for MATLAB and run it in a computer cluster, then it can even solve million DOFs problem in acceptable time.
The problem why not using MATLAB to build finite element analysis program is like what brad has said. In addition to brad's comment I want to say that a new FEA program written in MATLAB must undertaken first a long time of validation (bug fixing) and I don't want to spend time for this, so I will wait many years till many engineers have validated this new MATLAB-FEA program(spend their time), any suggestion?

cheers

 

[kreative]

Hi all i would like to add my 2 cents to the question why not MATLAB. i havent used matlab but i have written code in NASTRAN for some simple analysis. the first hurdle is to let the program know the shape of the model in terms of the co-ordinates of the nodes. i had to physically enter the node co-ordinate and link the right nodes to form the elements. this i think is another aspect that consumes time. one more thing. in FEA s/w animation to some extent helps in determining whether the constaraints are correctly applied. but if u look at a result file in data format its hard to say whether ur results are right. thanks

ym

 

[zuardy]

Hi folks,
don't underestimate the MATLAB capability concerning graphical interface. One could certainly program MATLAB for this purpose, it has atractiv graphic capabilities.
I have also program MATLAB to read (parse) the NASTRAN input & results file, write elements to input file and display the reluts graphically.
The important points, why not programmed MATLAB to a powerfull FEA-Program, are the long development time concerning validation and its relibility. I'm familiar with PATRAN/NASTRAN and ANSYS, so I love it very much : i love it because i know it!
So if the FEA-program, say ANSYS, would be translated in MATLAB later, it would not effect me as long as the GUI remains the same!

cheers

 

[tld23]

Guys,
talk about reinventing the wheel, what planet have you been on etc. Try looking at:

http://www.femlab.com/

and see what wonderful things you can do with MATlab

TERRY

 

[zuardy]

Terry,
have you overlooked my first post in this thread? I have namely mentioned the address. I even have found some more address related to MATLAB for FEM, but i'm not so antusiast regarding such problems / trade offs mentioned by brad and me

cheers

 

[tld23]

Zuardy,
Yes I did miss your post - sorry. I will be moving to earth shortly which should help with my problems.
Personally (but from only reading the literature) FEMlab would seem good for research or speciaist applications, but it is far from ideal for day-to-day analyses.
TERRY