The beam element that is compatible with the lower-order shell element is the two-noded element.  This element is only exact for a constant moment distribution, i.e., applied end moments.  Any load between the nodes will not lead to an exact solution.  Thus, even the body load due to self-weight or the distributed load from a shell will lead to approximate solutions.  If one wants to start doing a dynamic analysis then the situation gets even worse particularly if one needs to get accurate results for higher modes.  The only way to capture a reasonable approximation to the truth is to perform (often significant) mesh refinement.  

This is the status quo of most commercial finite element systems; low-degree elements that, as a result, provide low-fidelity solutions.  In order to ensure that a reasonable level of engineering accuracy has been recovered requires mesh convergence studies and in some industries this simply does not happen.  As such it is only the factor of safety or the material's ductility that saves the practising engineer from computer-aided catastrophe.

This issue can be put another way.  If mesh refinement is not conducted then the finite element stresses will not be in equilibrium with the applied load and which design engineer worth his salt is prepared to accept such a fundamental issue?

Just some thoughts...

Evidently, shell elements are optimal for discretization of panels as a parts of space structures  (fuselages, wings, coaches and so). As stiffeners the rods, plates and other monolith parts are commonly used and this motivates using of beam elements for their modelling.   

Stiffeners are having  larger length when compared to their cross sectional dimensions whereas panels have length and width comparable  with less thickness.Hence beam elements are used to model the stiffeners and shell elements are used  to model the panels.

Just because the beam elements generally allow to describe various cross-sections. Only the rectangular cross-sections for stiffeners can be modelled with the shell finite-elements. Combination of both beam and shell finite elements allow then more precise modelling especially in terms of connections for stiffeners with panels.

often for aeronautic applications to design the structures we need to modelize the internal components that are renforced by either beams of rods. we talk about "stiffeners" otherwise when we talk about panel it already includes these stiffeners FE. 

besides the reasons given by "Alexander Konyukhov" and "David Bassir" I want to point to another reason.

When shell elements used as stiffeners they will be in "in-plane bending situation", which means that membrane part simulates bending.

1- membrane elements are stiff elements and need fine mesh -> more computational time.

2- incompatibility due the absence of rotational d.o.f

Interesting question Nils and I wonder what your motivation is for asking it.  It would seem to me that beams are ideal elements to model such stiffening members and should work well with shell elements.  How the low-fidelity commercial conforming elements perform with this sort of problem and how many pit-falls they present to the practising engineer is, thought, another question...

To reduce FE model size and improve precision of the results as beam elements formulation would be very easy to handle in FE modeling and simulation. Consider if you want to model a fuselage without beam elements, how much effort and time you need to model all the beam with the given sections?

In fact, you can use shell element for both panels and beam-like stiffeners. However, for computational efficiency, beam elements (involving much less degrees of freedom) are normally favored without compromising the computational accuracy. 

I thank everyone for their contributions to the debate.

Here are some more comments from my side:

It might be difficult to use beam elements in case of composite stiffeners.

Nowadays, the number of DOFs is not a real bottleneck.

How many preprocessors are able to visualize cross-sectional shapes  of beam elements ? Verification of finite element models is becoming increasingly important.

I would second the idea that beam elements allow for a very easy redefinition of the cross section without a need for remodeling or remeshing, thus saving time.

Also, beam elements provide exact solutions with a minimum number of nodes and dofs. As pointed out by Mr Boutagouga, shell formulations (at least first order shells) do not handle bending (in-plane) very well compared to beams. If I recall correctly shell elements are prone to shear lockup under in-plane bending and tend to overestimate the actual bending stiffness. Moreover, even if this is not the case, and even considering a fine mesh and much increased computational efforts, they may not provide as good (or exact) a solution as a beam element would. Cook's 'Concepts and Applications of Finite Element Analysis' has interesting discussions on such topics.

As for composite beams, I think that a beam with homogenized composite properties may work well unless you want to make a ply by ply failure analysis. And even then, probably a multi-scale model would be more appropriate.

Finally, I would have to disagree on the number of dof's not being a bottle neck anymore. The selection of the right element type is not only a matter of computational effort, but also of precision. I am not sure that being able to solve larger models is a good reason to produce a model that is longer to make, longer to solve and has lower fidelity. Also,  the fact it would look better does not mean it would be better in terms of outputs.

Best regards,

Laurent

The beam element that is compatible with the lower-order shell element is the two-noded element.  This element is only exact for a constant moment distribution, i.e., applied end moments.  Any load between the nodes will not lead to an exact solution.  Thus, even the body load due to self-weight or the distributed load from a shell will lead to approximate solutions.  If one wants to start doing a dynamic analysis then the situation gets even worse particularly if one needs to get accurate results for higher modes.  The only way to capture a reasonable approximation to the truth is to perform (often significant) mesh refinement.  

This is the status quo of most commercial finite element systems; low-degree elements that, as a result, provide low-fidelity solutions.  In order to ensure that a reasonable level of engineering accuracy has been recovered requires mesh convergence studies and in some industries this simply does not happen.  As such it is only the factor of safety or the material's ductility that saves the practising engineer from computer-aided catastrophe.

This issue can be put another way.  If mesh refinement is not conducted then the finite element stresses will not be in equilibrium with the applied load and which design engineer worth his salt is prepared to accept such a fundamental issue?

Just some thoughts...

@Mr. Ramsey

Thank you for correcting me and for the pretty interesting comment on the importance of the (oft-neglected) mesh refinement and convergence analysis.