Discussion Forum

Hi

these can come from many issues: insufficient mesh density, to coarse interpolation elements, time sequence versus mesh density not respecting the Fourier number limits, numerical oscillations due to scaling or even smoothing issues in the postprocessing, many to choose from ;)

--
Good luck
Ivar

hi
thanks a lot. you mean I have to go for trail and error method to fix this .

Hi

not necesarily, you should check the usual first: what is the expected solution at t=0 and there after, does the mesh resolve the gradients, are all initial conditions "ogical" and reasonnable ...

The idea with FEM is to get a more precise idea of a asolution that you already know or feel, not to discover something fully unknown, as then you have no way to say if its correct or not, model validation and verification is part of the analysis process

--
Good luck
Ivar

hi ivar
thank you for your reply. I am doing like this. First import my text file in interpolation function(and plot it , which already shown to you). I use int1(t[1/s]) in the initial temperature field(means using temperature BC).I am using fine mesh only. In the interpolation filed I use linear, extrapolation field I use constant, in time dependant field I use range(0,1,24000)s.
and my input text file is attached with this and material is aluminium.

Apart from this where I am making mistake? Since I am new to comsol, I am feeling hard to find it.

Hi

int1(t[1/s]) corresponds to "t*volume" (if in 3D) or "t*area" (if in 2D) of your domain so this is a particular BC condition, not sure I understand it though.

Fine mesh is not enough as it's the mesh density in the required regions with high T gradients that counts, so even a "coarse" mesh could be perfectly OK, it depends also on your time steps, linked via the Fourier number Fn = alpha*delta_time_step/h^2 where h is the local mesh average size and alpha the thermal diffusivity alpha = k/rho/Cp of your "material"

With a time step of 1 seconds you will store 24000 solutions, that is a lot ! you will probably get a few Gb files.
As the general solution of a thermal flux case is often following an erfc() function exponential law, often you can use a log time step or a power time step i.e. 10^{range(-1,0.5,4.5)} or perhaps a 2^{range(-1,1,15)} (again it depens on the details of your model), use the thermal diffusivity to estimate the time to get to an equilibrium, check your thermal conduction reference books.

--
Good luck
Ivar

hi ivar
thank you for your suggestions.Its working fine with the 10^ and 2^ time steps. But I need more details about verification and validation of models. Can you tell me the guidelines to model a good physics ? Because I feel that comsol documentation is not in detail , for beginners like me its hard to get inputs from that.

Hi

Verification and validation of any model is "good (I would add mandatory) engineering practice", see i.e. try a google search of "NAFEMS validation verification" (www.nafems.org), they have a good document (and good courses to follow by the way)

The main rule, you should have a reasonable estimate of what you expect to get out of your model (any model), or at least a few points through which a specific parameter (global energy, given heat flux, teperature ...) should pass. That you know by hand calculations or an analytical solutions, or from another model (I use often Maple and MapleSim (www.maplesoft.com), a Modelica based tool (www.modelica.org), to get some good estimates of an even simpler model, so that my more general FEM model is verified).

It's worth to search for FEM model validation procedures, by grate instances such as NASA, ESA, NAFEMS ... "quality" documents, which are often good sources for "good practice" that has been tested in detail, even if, for simple models, you do not necesarily need to follow all steps

--
Good luck
Ivar

To add a bit to Ivar's last comment, best practices promoted by organizations like NASA, NAFEMS, etc frequently point to (at least) two verification steps: 1/ checking that the software you use performs on your machine the same way as it does on the developers' (This is frequently referred to as Software Quality Assurance, or SQA) and 2/ ascertaining that the software returns results that are consistent with established and trusted, exact or approximate, solutions for the same equations (This is referred to as Numerical Code Validation, or NCV). Frequently, they'll recommend that you demonstrate that the equations you chose to solve in the software are appropriate for the real-word situation; That is referred to as validation.

COMSOL recently added to our website this page that gathers well over a hundred verification and validation models and their documentation. Those models can be filtered by discipline or by module, as well as searched through via a free word search.

Best,

Jeff

-------------------
Jeff Hiller