Parametric study meshing strategy

[Jlog50]

Hi all,

I would like to discuss about the importance of meshing in parametric studies, I know that the mesh strategy has the utmost importance in producing good results, however, how important is it for parametric studies (especially ones which are non-dimensional, i.e. results are presented as a comparative study)? Anyone have any experience or advice on this they would like to share.

Thanks

 

[rb1957]

depends on what you mean by "parametric studies".

if you're just doing patch tests and want to see how the element behaves, then it probably doesn't matter.

if you're comparing different materials, then it's important, as you need to know the maximum stress.

 

[Jlog50]

Thanks for your input rb1957, by parametric study I mean varying one aspect of the design and doing a non-dimensional comparison between each design. For example "the stress reduces by a factor of 2 when the diameter is doubled". I hope that makes my question more clear.

Thanks again.

 

[rb1957]

so then you do want the peak stress, no?

so you want to know your model has converged.

i don't think you need to use the same mesh density for both models. and there are several ways to define mesh density ... absolute element size, element size/critical dimension, ... i think all you need to say is "this is the peak stress in the baseline model, and this is the peak stress in the modeified model"

 

[flash3780]

I would recommend overdoing it on the mesh if there's any doubt. Too coarse of a mesh can give anti-conservative results in stress/strain models. To check your results in a stress/strain model, you'll want to look at both the element stress solution and the nodal stress solution. If there's a significant difference between the two your mesh probably isn't fine enough.

There are also general guidelines to follow for stress/strain models: 3-5 elements through the thickness of a thin section, 5-10 elements around a radius, etc.

At the end of the day, you may want to refine the mesh on your optimized shape to see if the stress results change any. That will give you confidence that the solution is grid-independent.

 

[aidanpd]

I would echo rb1957 on this one and suggest you ensure that your model has fully converged. This can be done by using an iterative process of gradual mesh refinement until your peak stresses converge on some value.

I would be careful of over refining a mesh though, as stress singularities could significantly skew your results (i.e sharp corners, at application point of a point load, at regions of contact between two bodies, etc.).

Alternatively, there a are several resources available to aid you in your analysis based on the presence of common geometric features such as ESDU datasheets and textbooks (Roarks for example).

Best of luck...

 

[KevinDeSmet]

Even though it will take more time on an already more time consuming process (lots of slightly different models, lots of analysis runs) I would opt for adaptive meshing techniques. I guess it depends what software you're running but I believe that will give you a fairly accurate comparative convergence.

Certified SolidWorks Professional