According to Dr. Dhondt’s book “The Finite Element Method for Three-Dimensional Thermomechanical Applications”, ccx uses the α time integration scheme with predictor-corrector. In my own finite element programm, I found the predictor-corrector methods are not suitable for transient dynamic analysis considering geometric nonlinearity, such as rubber materials(large strain elasticity). During the calculation, the Jacobian of the finite elements sometimes has negative values, or need more iterations than the standard Newmark family of methods without predictor-corrector. I guess the predictor value may exceed the local convergence region of Newton iteration, resulting in divergent results.
Have you encountered similar situations when using Calculix?
Have not tried this in CalculiX yet, but from a self-written code I know that this happens easily even in quasi-static simulations. This is very common in general, as I have heard from multiple colleagues. If the increment is too large, there will be degradation of the elements.
I agree with Oliver,
Each curve represents the same analysis but with different cell/timestep ratio from poor to appropriate. Thermal transient without mechanical response should be solved first to be sure there is a clean temperature profile inside the elements close to large gradient areas.
Thanks for your reply. I will keep both formats and test them in more cases.