Does Altair Inspires "faster" optimization use tetrahedron of first order or second order?

Dogan_22589
Dogan_22589 Altair Community Member
edited April 12 in Community Q&A

Hello, 

the FE-analysis of Altair Inspire distinguishes between first order and second order thetrahedrons. (faster / more accurate ) 

Is it the same for the topolgy optimization ?  

If that is the case and Altair Inspires suggests than one should use the faster solution (first order thetrahedrons) for topology optimizations then how is it possible to compare pre optimization and post optimization FE analysis with each other if I used the more accurate FE in my pre optimization process.

The more accurate option for the optimization consistently fails to run on my PC.  

The part I am working on is large and thin which leads to huge discrepancies between faster and more accurate solutions. Therefore the optimized geometry does not fit to the actual load case. 

I hope my question is understandable. 

Thanks in advance.

Dogan 

Tagged:

Answers

  • Adriano_Koga
    Adriano_Koga
    Altair Employee
    edited April 12

    indeed you're correct. Faster is 1st order and 'more accurate' is 2nd order mesh.

    For topology optimization, specially in Inspire, usually I don't see any advantage on using 2nd order mesh to perform the optimization, as it will increase a lot memory requirements and solver time.

    Additionally, for topology optimization, for example, each element is just like a "pixel" for your model, and in the end the solver will "turn on or off" these pixels. Usually I prefer to use 1st order model, but refine more the model, to have more pixels to play with, and give the optimizer more freedom, than using 2nd order.

     

    1st/2nd order potentially have indeed a large difference in stress values, but stress is not usually so accurate in topology optimization, so I wouldn't mind too much about this.

    If your part is thin, maybe shell/surface model would benefit. Otherwise, whether using 1st or 2nd order tetras, you would need to refine the model to get decent results anyway.

Categories