AMS/DTNODA For Long Simulation Run Times

Crashphys
Crashphys Altair Community Member
edited October 2020 in Community Q&A

Hi all,

 

Just got through this video, which was incredibly helpful in establishing a procedure for using DT/NODA/CST:


 

I am currently working on something for which I hope to model approximately 20 seconds of simulation time, possibly more. The video has the following process:

1. Get the time step without any mass scaling

2. Use DT/NODA/CST with a Tsca of 0.9 and a time step of 1.2*Original Time step

3. Ensure mass error DM/M < 0.02

4. Keep going up by 1.2x until you see DM/M approach 0.02 (For safety, use 0.016)

 

My question is: You use the 0.016 value because mass error tends to increase over the simulation, and you don't know by how much. I've noticed most explicit analysis simulations run about 0.05 seconds, though maybe a bit longer for crash analysis. In anyone's experience, have they run a simulation that is something like 20 seconds and experienced a mass error increase much higher than 0.004 from start to finish? My simulations take hours to run, and I prefer not to wait to see how much error I end up with.

 

Tl;dr: Doe s simulation length affect how much room for error you should allow? Essentially, does mass error increase over the simulation linearly?

 

EDIT: I was also wondering - in the attached files I'm running a simulation with an appropriately selected AMS, and it doesn't seem to be affecting the time step or simulation speed in any meaningful way... I even purposely tried to put in a ridiculously large imposed time step of 200 seconds and even 0 seconds to try to break it, and it didn't change a thing. Can anyone help me with this?

 

Thanks

Unable to find an attachment - read this blog

Tagged:

Answers

  • Simon Križnik
    Simon Križnik Altair Community Member
    edited October 2020

    Hi,

     

    timestep can drop during simulation due to characteristic element length reduction (compressive deformation) or high contact forces (penalty contact formulation). If conventional mass scaling is used the added mass can increase to maintain the imposed timestep. This issue is most common in crashworthiness analysis.

     

    20 seconds is huge for explicit method. In quasi-static analyis we can go for time scaling i.e. applying the load more rapidly than static testing, if kinetic energy stays small (<5%) compared to internal energy. But I am not sure if this technique is applicable to sloshing analysis.

     

    Output the number of AMS iterations per cycle via /DT/AMS/Iflag - Iflag =2 may help to monitor 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. There should be only a few cycles where INTER is controlling the timestep. 

     

    Unfortunately I am away from my HyperWorkstation so I can not run the model right now.

     

    This thread might be useful:

    https://community.altair.com/community?id=community_question&sys_id=d8a6c47a1b2bd0908017dc61ec4bcb38

     

  • Crashphys
    Crashphys Altair Community Member
    edited May 2019

    Hyperman,

     

    Thank you for the information. By 'long run times' I meant the fact that I am running for 20 seconds. I actually ended up plotting error over simulation time of 2 seconds and got the following plot. It seems that it's likely to hit a high mass error after 5-8 seconds, so I suppose you are right. It is not practical to run for 20 seconds (in case anyone ever wants to find out).

     

    image.png.b9ba0ac2c5bf56d9300619d282f7353e.png

     

    I have seen your quoted thread before and followed the advice there. One thing to note is that Istf=4 is not compatible with SPH. I'm not sure why, but when I tried it, the contact interfaces didn't work correctly. I also confirmed this through the RADIOSS User Guide. You can only use Istf=0 or 1 with SPH.

     

    For number of iterations, I am hovering around 7-10. In terms of iterations, here is what I have in the _0001.out file:

     

       CYCLE NUMBER13000 TOTAL C.G. ITERATION NUMBER=    7 RELATIVE RESIDUAL NORM= 0.6068E+10 REFERENCE RESIDUAL NORM 0.9346E+10
     13000  0.6887      0.5176E-04 NODE      428921   0.8%  0.2817E+08  0.2279E+06   4129.      0.2818E+08   0.000    

     

    My imposed time step is 5X what I got from determining it from DT/NODA/CST, which I set as DT= 2.22E-4 in AMS

     

    It doesn't seem that INTER is controlling the time step, and my energy error is quite reasonable up to a certain point. Am I interpreting the above information correctly? Am I correct in assuming that there is good convergence up until a certain run time in this model?

  • Simon Križnik
    Simon Križnik Altair Community Member
    edited May 2019

    It is a good sign if the energy error is reasonable and C.G. iteration number stays below 30. However, it is recommended to also run conventional mass scaling and compare displacements, stresses and strains.

    Unable to find an attachment - read this blog