Hi Gianmarco,
I found the problem. It is purely numerical. Your mismatch with Seismostruct happens in the elastic range, and this was very strange to me... But after some checks, I found that the use of the norm of displacement increment test with a large time step is causing the acceptance of a solution which is quite far from the equilibrium path in the elastic range. You were using a large tolerance (1.0e-4) but I even tried 1.0e-9 and nothing changed. This probably means your models has a stiff system.
Do the following:
- Use NormUnbalance test (tol = 1.0e-5, iter = 100)
- Change from Penalty to Transformation (Penalty does not work well with NormUnbalance)
- With Newton or NewtonWithLineSearch it takes about 20 iteration each step, try KrylovNewton. It is the one that gives the best convergence in this problem (about 5 iterations)
Note 1: In this case use can safely use Transformation instead of Penalty as constraint handler because there are no strange or circular dependencies in the constrained nodes. I understand that in some complex models the relation between constrained nodes can be quite cumbersome, and so the transformation may not work.
In those cases you are forced to use Penalty, and thus a displacement test (the unbalance cannot be used with penalty). When using a displacement test with this kind of problem, use smaller time steps.
Note 2: With this new configuration, Linear or PDelta transformation give almost the same result (displacement are not so large), with PDelta showing just a lower strength in the nonlinear range. However I noticed something interesting. When I tried Corotational transformation (which is the most general and complete kind of nonlinear kinematics in OpenSEES for beams), the structure was twisting about the Z-axis. This happend because of the triangular discretization we use in the fiber section, because with those few fibers, fiber where distributed in a not-so-perfectly symmetric way, thus causing the twisting of the structure when corotational method was used. This did not happen with the quad mesh, which is instead symmetric. So we are currently implementing the quad fiber discretization in STKO, at least for geometries that allows it.
Hope this helps
Massimo