4. Practical aspects of use

This method is used automatically as soon as the chosen behavior is not available in plane constraints, for C_ PLAN or shell-type models: COQUE_3D, DKT, TUYAU. In practice, this (automatically) increases the number of internal variables in behavior by 4.

To converge well, it is advisable to update the tangent matrix (if possible, at all iterations: REAC_ITER = 1, or all \(n\) iterations, with \(n\) small).

This method therefore allows a great flexibility of use in relation to behaviors: it is sufficient for a behavior to be available in axisymmetry or in plane deformation for it to also be usable under plane stresses.

As with all integrations of non-linear behavior models, it is strongly recommended to give a small convergence criterion (leave the default value at \({10}^{–}6\).).

The advantage of the modified approach is a better satisfaction of the plane stress condition at each Gauss point (\(\mid {\sigma }_{\text{zz}}^{\text{mod}\text{if},(n)}\mid \text{<<}\mid {\sigma }_{\text{zz}}^{\text{orig},(n)}\mid\) for each \(n\)). In some cases, it is essential to converge a calculation, in particular for softening laws of behavior.

On the other hand, because of an additional loop, the modified procedure is more expensive, especially because the loop includes the call to the « 3D behavior law » module. However, the additional cost due to heavier local calculations can be offset by a gain in the number of global Newton iterations, which is usually lower for the modified algorithm. This gain in the number of global iterations is not guaranteed, which means that the additional iterative loop is not activated by default (ITER_CPLAN_MAXI =1). It has also been observed that as long as you choose ITER_CPLAN_MAXI > 1, it is preferable to use ITER_CPLAN_MAXI > 5.