yade-users team mailing list archive

[Bruno Chareyre <bruno.chareyre@xxxxxxxxxxx>]

>>> Bruno, how does this value compare to critical timestep calculated from
>>> elastic wave propagation speed along the particle calculated for
>>> cotinuum, i.e. , which is 2*particle diameter / sqrt(Young's modulus /
>>> density)?
>>>       
> ElasticBodyParameters::young, from which stiffnesses of elastic links
> are normally calculated.
>   
Ok... So I really don't know how the two values compare. But I'm 
interested by the result of the comparison.

>> There is a simple way to evaluate if a timestep is good or not :
>> run the simulation and watch the evolution of unbalanced force (now 8th 
>> column in WallStresses files generated by TriaxialStateRecorder) while 
>> you increment dt manualy. When unbalanced force start to tend to a 
>> constant non-zero value, or even increases more and more, dt is too large.
>> You will see if your dt could be much larger without any trouble.
>>   
>>     
> That was not meant for triaxial test,
The fact that dt is less than the critical value is not related to the 
simulation (TriaxialTest or anything else).
The point was just that TriaxialTest is one way to test if an equation 
for computing dt_crit is correct or not.
If your equation is correct in TriaxialTest, it will be correct for 
almost any simulation, and vice-versa (because TriaxialTest includes all 
possible states, from granular gaz to very dense packing, which means 
variations of in a wide range of values).

> I know I can use unbalanced force
> there. (I will probably move that code somewhere else to make it
> possible to use it in different simulations.)
>
>   
Good point. We also discussed with Janek a way of computing unb_f 
without additional loops : the mean contact force and the mean forces on 
body could be computed respectively in interaction law dispatcher and 
position integrator, or something like that.
The unbalanced force could even be displayed in GUI with the current 
timestep as it is very usefull in most simulations.

> The advantage is that it makes a (very rough) estimate without ever
> running or even creating the simulation.
>
>   
I see. But if it's independant on the density (i.e. number of contacts), 
it will be either far too small in some situations or too large in 
others. I think ElasticCriterionTimestepper was coded with the same idea 
in mind. Not sure it was a big success since people always got dt values 
that were so small that it was like nothing was happening in the simulation.
The only situation where GlobalStiffness is not relevant is when you 
have high speed impacts (problems with high initial velocities) , in 
such case dt_crit depends on the ratio (size of bodies / initial 
velocity) because you don't want bodies to pass through each other in 
one iteration. Still not related with elastic waves...


> V.
>
> _______________________________________________
> Yade-users mailing list
> Yade-users@xxxxxxxxxxxxxxxx
> https://lists.berlios.de/mailman/listinfo/yade-users
>
>   


-- 
 
_______________
Chareyre Bruno
Maitre de conference

Institut National Polytechnique de Grenoble
Laboratoire 3S (Soils Solids Structures) - bureau E145
BP 53 - 38041, Grenoble cedex 9 - France
Tél : 33 4 56 52 86 21
Fax : 33 4 76 82 70 43
________________

_______________________________________________
Yade-users mailing list
Yade-users@xxxxxxxxxxxxxxxx
https://lists.berlios.de/mailman/listinfo/yade-users