← Back to team overview

yade-users team mailing list archive

Re: Periodic triaxial

 


    And in fact the rate of loading is what you'll find in Thornton
    and Cundall when defining the relative velocity across periods, it
    is my best bet before reading the papers.

Yes, supposing a strain-controlled test you will find the usage of the strain rate of loading to update relative velocities. Maybe is that what we need once adopting ScGeom?
Yes, this is what we need. Note that updating velocity should be done on bodies themselves to be really correct (i.e. there is really no "edge" of the packing, what is supposed to be obtained with homotheticCellResize>0). Updating relative velocity for interactions across period is another independant problem.

I'll keep posting changes in periodicity in this thread, as I fix stuff.

For today, I just commited (r2249) those changes, which seems to give stable packings (before, it was exploding). I think it makes things correct regarding homothetic resize. See point (4) for the interractions part :

1 - stress definition in PeriIsoCompressor was wrong for ScGeom (sign mistake due to double-reverse);

2 - applyForceAtContactPoint cannot be used in periodic + ScGeom, since this function needs the wrapped position of each body (in Dem3Dof, wrapped position is an attribute, in ScGeom there is no such attribute and we don't want to define it as it costs cpu). The function call is replaced by separate applyForce/applyTorque using a "branch" computed using radius, normal, and penetration depth.

3 - Implement homothetic=2 properly. The acceleration of the period (i.e. change in velGrad) is applied on particles as an equivalent impulse. If velGrad is constant, there is no effect on particles. If velGrad changes, each particle is accelerated with (delta_velGrad*position). This is like giving initial velocities to particles so that they move homothetically.

TODO :
4 - Define relative velocity across period correctly. It should be done adding a velGrad*shift2 term, but shift2 is unknown in Law2... I'll maybe exploit the "shear" attribute of ScGeom to precompute a shifted shear disp increment in Ig2, and use it later in Law2. (btw : isn't that an interesting option to merge ScGeom and Dem3? Letting shear be the total OR incremental displacement?)

Cheers.

Bruno








Follow ups

References