Hi Hossein,

the basic reason is that, the properties of the elements based on DOF....

Some informations :

http://ansoft.com/staticassets/ANSYS/staticassets/resourcelibrary/confpaper/2006-Int-ANSYS-Conf-22.pdf

Hi Hossein,

one way to improve the results of solid element to agree with shell element is to divide the thckiness into even no of elements (eg. 2, or 4, or 6) cause solid element dosen't have rotation DOF.

Good luck

It depends on what elements you are using and the aspect ratio between the length in X-Y with length in Z (thickness) which is very important in determining what to use shell or solid elements. Therefore accuracy of solids is usually sensitive to mesh distortion and large aspect ratio (when shear deformation is known to be small)

my answer was based on the assumption that formulation of both shell and solid element is based on mindlin plate theory and use same order of interpolation function and with reduced integration scheme

The solid elements will tend to better capture any stress concentrations, since they can adapt to the geometry more precisely.  If the shell element cannot capture the geometry precisely, it may miss these stress concentrations.  Conversely, the solid element may be modeled with a geometric singularity (which may not actually exist), which would result in an artificially high stress.  Mesh refinement can exacerbate this situation.  In addition, shell elements will be based upon a particular shell theory, and inherent in such a theory are assumptions on behavior that are enforced in the element itself.  Solid elements tend to be based on the exact theory, such as theory of elasticity, and may not have such assumptions that shell elements would have.  For example, beam elements based upon basic beam theory do not include shear deformations.  For most cases, this would be an acceptable approximation (plane sections remain plane).  However, if the loading conditions produce high shear stresses (such as for short, deep beams), this approximation is inaccurate since the shear deformations are not insignificant.  In such a case, the beam elements will fail to work properly.

As Raafat mentioned, you also must consider the order of the interpolants and integration scheme.  This should be considered after you have identified the limitations from the theory itself (shell theory vs linear elasticity).  As you can see, there are multitude of sources for the differences.

Thanks for all replies.

Dear Raafat,

I tried using even number of elements through the thickness but the difference still exists. I also tried both reduced integration and not reduced elemets. I also tried linear and quadratic elements but the difference still exists.

Dear Jeff,

This difference exists in a simple models too. Consider a cantilever plate 1*1*0.05 m which can be modeled with both shell and solid. Apply a load such as gravity or sth else. Maximum stress in the fixed end differs from shell to solid regardless of the number of elements through the thickness for solid model.

Dear Auday,

Such a plate I mentioned can be meshed wth qualified elements with aspect ratios less then 4 or 5. (structure mesh in abaqus)

Dear Brian,

According to what you said, we can conclude that stress in solid model is lessreal and engineering design should be based on shell. Is that true? Which modeling should be trusted?

The second issue is about shell elements. There are 6 DOFs at each node. What do you think if 3 rotational DOFs can be removed and make a shell with 3 translational DOFs at each node? Do you think this can make the results differs from each other more severely? I couldn’t do this in Abaqus…

please describe specific example then we can advise you based on its behavior. you should know that each element has different formulations and with different advantages. when you solve a problem having transvers shear effect, not all element can consider its effect with same degree of accuracy. modeling different geometric and support and loading conditions needs also especial handling.

Finite element modeling is a numerical method, so we judge its results within certain tolerance of accuracy. 

Hi Hossein Soleimani,

let me highlight some points, that may be helpful in addition to previous answers.

As your analyze the results based on shell and solid models, you should compare stress resultants rather than local stresses. The classical shell (plate) theory in the sense of Mindlin or Reissner is not designed to reproduce local stresses accurately. So compute the stress resultants (forces and moments) from the stress distributions obtained by solid model, the values should agree better with those based on the shell theory.

The next point is related to boundary conditions. Assume you have a clamped edge, for example. If you use the shell theory, you set the displacements and rotations zero. However, there are many different possibilities to model clamped edge applying the solid model. All these 3-D models of clamped edge would lead to different stress distributions near the boundary (boundary layers).

Dear Hossein Soleimani,

As per my understanding the results should remain same if you apply in plane loading.

This problem will occur if you apply a transverse force or out of plane loading. Due to shear locking stiffness of the geometry meshed with solid elements will be over predicted and you will get wrong results. Better to use shell elements in such cases.

You may also find this link useful.

https://www.quora.com/What-is-shear-locking-in-FEA

Regards

Naveen