See also: http://ftp.lstc.com/anonymous/outgoing/support/FAQ/implicit.dynamic_relaxation http://ftp.lstc.com/anonymous/outgoing/support/FAQ/quasistatic http://ftp.lstc.com/anonymous/outgoing/support/FAQ/preload.tar _________________________________________________________________ RE: Explicit dynamic relaxation In explicit dynamic relaxation, DRTERM may never be reached, and in fact, DRTERM should not be specified. A check for "convergence" is made every NRCYCK time steps and if the convergence criterion is met, the DR phase is done. The convergence criterion is controlled by DRTOL and how fast it reaches that criterion is affected by DRFCTR . The output file "relax" reports a history of the convergence behavior and data in relax can be read and plotted by LS-PrePost. The load applied during DR should ramp from zero to the full value of the load and then that full load should be held constant for at least as long as the DR phase lasts. Furthermore, you want to make sure the ramp time of the load is shorter than the time it takes the DR phase to converge. jd 5/11/21 (no ticket) Some examples illustrating explicit dynamic relaxation (DR): Stress and deformation is initialized in a spinning blade using DR: http://ftp.lstc.com/anonymous/outgoing/support/FAQ_kw/blade.dr.k Shell part is stretched during DR phase and then is impacted in transient phase: http://ftp.lstc.com/anonymous/outgoing/support/FAQ_kw/explicit_dr.k Bolt preloaded during DR phase using thermal loads: http://ftp.lstc.com/anonymous/outgoing/support/FAQ_kw/bolt.expl_dr.k Bolt preloaded during DR phase using *initial_stress_section: http://ftp.lstc.com/anonymous/outgoing/support/FAQ_kw/bolt.initial_stress_section.4not1.k.gz (NOTE: When using drdisp.sif data from an *initial_stress_section DR run in an initialization by prescribed geometry (IDRFLG=2), include *initial_stress_section and define the corresponding curve with IDRFLG=0 and the preload stress at full value but very briefly (~ 1 time step). Ticket 2017061310000076) Initialization due to gravity using DR: http://ftp.lstc.com/anonymous/outgoing/support/FAQ_kw/gravity_preload_by_dr.k.gz Blade initialized in SPH bird vs. blade simulation: The first model is bird_blade_dr.k. A rotational body load is applied to the blade in the dynamic relaxation phase. Initlal velocities are applied in the subsequent transient phase in which the preloaded blade strikes the bird. At the end of the dynamic relaxation phase, a drdisp.sif file is written. This drdisp.sif file is used to initialize the blade in the second model very quickly in 100 time steps. The second model is bird_blade_dr_idrflg2.k. When you run this model, include "m=drdisp.sif jobid=run2" on the execution line. Please run the two models and compare the results of d3plot to run2.d3plot. jd Ticket#2019060610000102 _____________________________________________________________________ Q: After the DR phase completes and I'm not satisfied that the solution is close enough to steady state, can I restart the DR solution with a tighter tolerance or otherwise let it run for a longer time? A: Yes. Submit a small restart using the d3dump written at the conclusion of the DR phase and a small restart input deck that includes *control_dynamic_relaxation with the modified DRTOL or DRTERM value. Or, if you don't want to modify the DR analysis and just want to immediately start a transient run that initializes from the final state of the DR run, omit *control_dynamic_relaxation from the restart input deck. ________________________________________________________________________ Q: If I want to use only one curve for dynamic relaxation and transient, can I give the same constant load both for dyn. relaxation and transient run using SIDR as 2? A: Yes, that will work in many cases but a ramp in the dynamic relaxation curve is preferred as it will induce less dynamic response and thus the dynamic relaxation phase will converge faster and be less likely to overshoot the real preload stresses. You want to minimize this 'overshoot' of stresses as it may take the material into a nonlinear regime which might otherwise not be breeched. ___________________________________________________________ Q: Dynamic relaxation is slow to converge in my model. Is there any way to speed it up? A: It's tricky and often iterative to attain efficient convergence because the convergence of an explicit dynamic relaxation solution is affected by not only ramp time of the load, but also time step size and the natural frequency of the system being modeled. It helps to bear in mind that the "damping" applied by DR is simply a scaling down of nodal velocity every time step and convergence is judged based on the current "distortional" kinetic energy as compared to the high tide value of "distortional" kinetic energy. (Ticket#2018021310000134) By postprocessing d3drlf and the ASCII relax data, we can determine if it's reasonable to terminate the dynamic relaxation phase prior to the default, distortional-kinetic-energy-based convergence criterion being satisfied. In other words, plot time histories from the d3drlf database (*database_binary_d3drlf) to see if a solution reasonably near steady state is attained,i.e., displacement and/or stress time histories have "leveled off" to near constant values. You can do this postprocessing while the job continues to run. If you decide it's OK to end the DR phase, make a note of the time in the DR phase, and issue an "sw1" sense switch to stop the run and write a d3dump01 file. Submit a small restart using a restart input deck that sets the appropriate DRTERM in *control_dynamic_relaxation. An easier approach to stopping and restarting the analysis is to issue sense switch "SWE", that will cause the explicit dynamic relaxation to immediately terminate and the analysis to progress to the regular transient phase. Plotting velocity vectors using the d3drlf database may help to isolate the region where kinetic energy is not reducing in a timely manner. You can then focus on this region in seeking a remedy, e.g., by including contact damping (VDC). If DRFCTR is too small, deformation may be overly inhibited A symptom of this is if the "convergence" time history (plotted using the ASCII output file "relax") has bottomed out at too high a value. On the other hand, the "convergence" value decreasing very, very slowly so that convergence takes too long may indicate the DRFCTR is too high (too close to 1.0). Slow convergence in explicit DR is especially common in large structures with low natural frequencies. This is because it takes a long time for the structure to respond in its fundamental mode. Depending on the size of the explicit time step, convergence of dynamic relaxation could take millions of time steps, if it converges at all (overdamping may be an issue). A good alternative may be to use implicit dynamic relaxation. See http://ftp.lstc.com/anonymous/outgoing/support/FAQ/implicit.dynamic_relaxation For this approach, you should use a double precision executable of LS-DYNA. See also http://ftp.lstc.com/anonymous/outgoing/support/FAQ/implicit_guidelines . Since none of the common ASCII output files, e.g., glstat, matsum, elout, are written during DR (see bugzilla 14646), it may be helpful to set IDRFLG to -1 at *CONTROL_DYNAMIC_RELAXATION and specify *DATABASE_BINARY_D3THDT. This will generate a binary file, d3thdt, that you can open in LS-PrePost using File -> Open -> Time History File. In order to see nodal or element data, you will still need to select these with *DATABASE_HISTORY_option.