Learning Forum

[C Deho]

Hi,

I'm writing a code in MATLAB to calculate strains of a model using the 10-node tetrahedron element, SOLID 187. Using the shape function, I multiply the strain-displacement matrix B by the nodal displacements for each element to obtain the strains at the Gauss points, and then extrapolate these strains to the nodes.  While the strains generally match well for simple models, the error increases as the complexity of the model geometry increases. Decreasing mesh size does not lead to significant changes for the same geometry.

 

For a simple ring model, there were about 500 nodes where the error in the von Mises stress exceeded 5%, and as the model becomes more complex, the number of nodes exceeding 5% increases.

Using PRESOL, loci to verify the Gauss points, I found a significant discrepancy in the x-direction between the Gauss point locations calculated in MATLAB and those expected. The other directions match well. Do ANSYS internally use different methods to calculate gauss points for complex geometries?  Could there be a reason for this issue?

[Aurojyoti Prusty]

Hi,

Is there a discrepancy in the location of gauss points or the stress results? Generally PRESOL gives unaveraged element results which are discontinuous across element boundaries. What you have done is you did some sort of stress averaging in Matlab. For simple models this works fine because there is not much variation in strain between gauss points, but as the geometry changes significantly, the displacement gradient changes fast and averaging strains over there gives little more error than simple geometry cases. Please compare averaged stresses in simple model in Matlab and Ansys. Hope this helps.

Thanks,

Aurojyoti 

[C Deho]

There are discrepancy at both Gauss point locations and strain values. The Gauss points are derived using shape functions, and the strain-displacement matrix B, which is used to calculate strain, is obtained by differentiating these shape functions. Therefore, I suspect that the primary source of discrepancy lies in the Gauss point locations. I didn't used PRESOL command when I compare MATLAB result and ANSYS's.  I have also confirmed that the "Integration Point Results" option is set to "averaged."

                                           [simple model]

                                       [complex model]

I noticed a substantial discrepancy in the x-coordinate of integration points as the complexity of the model increases. This discrepancy does not seem to rely on the number of elements, as the results align well for simpler models.

Is there an option within ANSYS to disable such automatic adjustments to the Gauss point positions or shape functions? Or, if you have any insights or suggestions on how to address this issue, I would greatly appreciate your advice.

[melanie miller]

You used Matlab to do a stress averaging of some kind. Because there is minimal fluctuation in strain between gauss points for basic models, this works well. However, when the geometry varies greatly, the displacement gradient changes quickly, and averaging the strains throughout that region results in slightly more inaccuracy than in situations with simple geometry dash lite.