Recommended Checklist
True incompatible kinematic conditions (IKC) must be carefully checked and resolved when occurring. AMS may either fail or run with a very low target time step, leading to poor performances, whereas the classic time step control for the same model can run normally. In that sense, AMS can reveal modeling weaknesses.
Any targeted model for AMS application should run first in /DT/NODA/CST with a reasonable energy error (ERROR < +2%) and acceptable added mass (MAS.ER < 0.02) along its simulation time. A model unable to run with /DT/NODA/CST will not run with AMS either.
Too much mass added when using /DT/NODA/CST can cause result differences when compared to an AMS result where no mass is added. If necessary, rerun the /DT/NODA/CST model with a lower time step to reduce the amount of mass added and then compare the AMS results.
An AMS job may either fail or diverge, most likely with an error message. However, once an AMS run goes through, several aspects remain to be carefully checked in order to insure a maximized elapsed time saving, as well as a reliable result quality.
- Time step value evolution and comparison with the targeted time
step:
- The time step should follow the targeted time step without time step drop (occasionally)
- If the time step is at any time smaller than the targeted one, the target time step should be reduced
- Entity controlling the time step at cycle (SOLID, SHELL, SPRING, BEAM, NODE, INTER, and so on.)
- Energy Error evolution (should decrease slowly negatively - not increase)
- Output the number of AMS iterations per cycle via
/DT/AMS/Iflag - Iflag =2 may help monitoring convergence quality at no
extra CPU cost.
Maximum allowed iterations before the divergence stops is 1000. 75 to 100 iteration per cycle is a sign of a poor convergence 50 still may provide some speedup
30 iterations or less is considered a good convergence.
Basic Quality Indexes
With constant nodal time step control (/DT/NODA/CST), the energy error (ERROR %) and added mass (MASS.ER) are the first ones to check inside the output file (_0001.out).
With AMS, the added mass is irrelevant. Review the energy error, the time step variation, and the number of iteration per cycle remain.
Common to both approaches are the energy balance curves.
Results Accuracy
The numerical effect of AMS on results is similar to lowering the highest eigen frequencies of the structure. Since AMS affects high frequencies, it is suitable for quasi-static, drop test, and manufacturing (stamping) simulations. It is not recommended for highly deformed structures or high velocity phenomena, like explosion and ballistics simulation. Occasionally, AMS may be used in crash simulation, if selectively applied to a group of finely meshed structural parts, potentially penalizing the computing performance of the entire model. In such cases, the AMS target time step of these selected parts (excluding safety subsystems, such as: barriers, dummies, airbags, and restraint systems), must be set to the time step value of the rest of the model where a classical mass scaling is applied.
In most cases, the buckling behavior is not affected by eigen vectors (then by AMS). In specific cases like box beam crushing, the width and amplitude of peak forces may be altered since the peak force may not be detected, due to an AMS higher time step. The buckling pattern is correct, but the absorbed energy is not the same on the first peak.
Rigid bodies rotation: rotational inertia of small rigid bodies may be affected by AMS. Inertia of these small rigid bodies will be increased in order to achieve the target time step.
When comparing AMS results with reference run, make sure the added mass in the reference run is low enough so the results are not affected.
Lessons Learned
- Too close to the traditional constant nodal time step value, the AMS numerical stability is excellent and results almost identical to the "reference". Of course, speedups are poor due to AMS iteration cost. In theory, a targeted x3 factor will return an "effective speedup" of x1 which is of no benefit.
- Slightly above the previous case AMS numerical stability remains satisfactory mostly showing references to element related minimum time steps (possible references to interfaces). Results are still close to the "reference", AMS iteration cost becomes amortized and speedups start to rise. As an estimate targeted factors from x3 to x10, most likely return effective speedups ranging from x1.5 to x3, possibly more.
- Above that, numerical stability may look satisfactory with possibly, higher numbers of iterations per cycle and more references to interface related minimum time steps. Results may either still look acceptable or significantly different compared to the "reference". Speedups continue to grow, unless penalty-based interfaces dictate the minimum time step causing iteration number increase and possible time step drops, what affects the speedup. However, targeted factors from x10 to x20 possibly x30 may return effective speedups ranging respectively from 4 to 6, possibly 9 (more was found in stamping simulation) that should NOT be considered a success as long as AMS results are not checked and compared to above defined "reference" results.
- Above previous cases, computations either diverge then stop with an explicit AMS message or will stop due to Energy Error limit (no added mass) or the time step dramatically drops and the run needs to be interactively stopped.
The 1.out File Examples illustrates the AMS effect on the Radioss output file (1.out) depending on the entered target time step.
Hints
When models contain parts with very different mesh sizes it may be better to only apply AMS to the parts penalizing the time step. Then, for optimized computing time performances, it is advised to also apply a classical mass scaling to the parts not belonging to /AMS part group, otherwise, none AMS processed parts are computed, by default, in natural Element time step.
/DT/AMS/1
Tol_AMS
/DT/NODA/CST
It is pointless to use a higher time step in AMS than the one used in classic mass scaling, since the smaller time step limits performances. If no part group is specified (blank line) or is equal to 0, then AMS is applied to the model in its whole, and adding /DT/NODA/CST is irrelevant.
For AMS, like in standard mass scaling, it is recommended to not have friction in a TYPE11 contact if a TYPE7 (already handling friction) contact is already defined for the same parts. This avoids drops of time step and helps model convergence. This recommendation is obsolete, if TYPE11 friction is using Iform=2 available for edge-to-edge contacts since 13.0.
** WARNING SECONDARY NODE OF AN INTERFACE TYPE2 & AMS INTERFACE TYPE[7 or 11], ID=xxxxxx: SECONDARY NODE ID=yyyyyyy IS ALSO SECONDARY NODE OF AN INTERFACE TYPE2 THE NODE CONTACT STIFFNESS WILL BE DE-ACTIVATED CASE OF /DT/AMS
This contact deactivation can be avoided by using Spotflag=25, 27 or 28 (TYPE2 penalty formulation).
Kinematic formulation (TYPE2) for spotwelds may alter AMS performances when there are many spotwelds, particularly hexa spotwelds.
Some dynamic cases, often in elastic state, exhibit strange elastic vibrations (showing arlequin-like von Mises contours) forcing to lower the AMS target time step leading to poor AMS performances. These vibrations can be reduced and the AMS target time step re-increased for better performances by applying Rayleigh damping. The recommended damping value is: