When I attempt to do a springback analysis, I sometimes have difficulty in obtaining a converged solution. What options are available to improve convergence of springback?

There are several ways to improve the convergence

a) Use multiple steps in implicit analysis. The step number is defined by NSTEPSB (the fourth parameter under *CONTROL_IMPLICIT_GENERAL). For most of problems, one step will be enough. However, for a part that is very flexible, it can be difficult to get a converged solution. In this case, a multiple step approach is required.

b) Change the value of SCALE (the second parameter of *CONTROL_IMPLICIT_STABILIZATION) for multiple steps springback analysis. If during the first step it is difficult to get a converged solution, then use a larger value; if the last step is difficult, they try a smaller value. If both first and last steps are difficult, then try more steps.

c) Change the value of LSTOL, (the seventh parameter under the card of *CONTROL_IMPLICIT_NONLINEAR). In some case, it is better to use small value; while for the others, larger is better.

d) Use a different solver LSOLVER (the first parameter under the card of *CONTROL_IMPLICIT_LINEAR). It has been found that the different solvers have very different convergence behaviors.

e) Try constraining different nodes to remove rigid body motion as both the nodes and constraints can have a large influence in obtaining a converged solution. As long as the constraint is properly applied to eliminate all rigid body rotation and translation modes, the final results will be very similar.

f) If the implicit analysis is terminated because of negative eigenvalues, then use the default value of NEGEIG (the third parameter of the card of CONTROL_IMPLICIT_LINEAR) which causes the solver to ignore the negative eigenvalue and continue the analysis.

g) Check the mesh for mesh distortion which might have been introduced during the forming or trimming operation.

h) Check the aspect ratio of the elements, too large a value of this ratio will cause convergence to be very difficult.

i) Coarsen the mesh. It has been found that it is not necessary to use a very fine mesh during the springback analysis. It is possible to remove the unnecessary elements from the original fine mesh that was obtained from the forming analysis through “coarsening”. When a very fine mesh is used for springback analysis, it is much more difficult to get a converged solution.

j) It will help to get a converged solution if the parameter IRNXX=-2 (under the card of *CONTROL_SHELL)