yade-dev team mailing list archive

[Bruno Chareyre <bruno.chareyre@xxxxxxxxxxx>]

Wow! Man, it will be hard for me to improve the existing code if you 
revert my changes (I need to recompute Hsize in many places atm because 
you didn't want it in Cell...). Did the change affect your simulations, 
they should not?
I should go back to the copy/paste method to be able to keep code in 
order for at least a few days? ;)

> Wading through the code a few things I need explanation as I am
> otherwise thinking of reverting them:
>
> 1. Why use logarithmic strain? I have code for plotting strain-stress
> curves, for instance, which uses PeriTriaxController; those are plotted
> (at least for concrete) with linear (small) strain. Is there some
> advantage besides that it is customarily used in geomechanics? 
>
>   

If it causes trouble, just tell me and I'll see how to fix it. It was 
not supposed to change anything for you (especially if you work in small 
strains, you should not see any difference). So why do you need to 
revert this?

This is not geomechanics, just "correct" mechanics. It makes things 
consistent.
- Take real a steel bar and make a tensile test on it. Plot force vs. 
dL/L, it is not linear (even if it looks linear for small strains). Plot 
force vs. log((L+dL)/L) , it is always linear (provided you didn't enter 
plasticity ofc). The same happen in DEM simulations.
- volume change : with your definition ev!=(e11+e22+e33), with Log, they 
are always equal.
- apply a strain rate of -10e-2/sec for 50 seconds. Result with your 
strain 0.393 (not equal to strainRate*time!), result with log : -0.5.
- with log, there is no upper limit to strain (except the grain 
size/period size condition), you can compress 100, 200, 300%. With your 
definition, e=100% has no meaning.
- most users won't notice any difference if they are in small strain, 
you don't need to adapt anything or "think" about it. When "your" strain 
is 0.01,   log(1.01)/1 gives 0.0995 (log is just slightly more 
accurate). And if you don't remove the exponential, target will be 
reached exactly.

> 2. Why are you removing refSize everywhere? The initial cell is always a
> box (perpendicular), and refSize has clear meaning. Current size also
> has meaning, and is needed by the collider.
>   
I'm not removing refSize at all, and I agree on the clear meaning of it 
(size is different, see #1).
Did it change something for you?

The current (reverted) code is doing approximately this :

maxGrow = dt*maxRate*refSize;(in meters)
Grow = targetStrain*refSize;(in meters)
compare(Grow,maxGrow);
velGrad = Grow/refSize;

Does it make sense to multiply THEN divide by refSize?!
I suggested this, simpler and generalised more easily :

maxGrow = dt*maxRate;(in [-])
Grow = targetStrain*refSize;( in [-])
compare(Grow,maxGrow);
velGrad = Grow/dt;

would have been even better :

Grow = targetStrain*refSize;( in [-])
velGrad = Grow/dt;
compare(velGrad,maxRate);




#1. I'm currently working with the polar decomposition of F to define 
proper stretches and rotations. It is the only way to define a "size"  
without rotational artifacts for e.g. simple-shear (note that simple 
shear is the most interesting non-axis-aligned loading, pure shear makes 
no sense since it is just a classical triaxial test performed on rotated 
box).

It almost work, and it allows shearing the cell forever without reaching 
the "2*halfsize>size" limit. It simplifies many parts in the code 
(collider using period of lengths refSize instead of getSize...). Should 
I commit this to trunk without copy/paste? You will not revert it the 
day after? :)

> (And really please don't put so many spaces to code (and not to code you
> copy&paste), I admit this is not rational, but I have hard time reading
> spacey code and screen cannot take as much code as it could.)
>   
What you see in my code is not specially "french" (or I missed 
something), it is the default c++ indentation of kdevelop (yes, I auto 
indent...). Sorry for that.
I'm, looking forward for Anton's guidelines in order to set kdevelop 
indent mode. There should be a way to commit this setting in 
trunk/yade.kdevelop.

Bruno

--

_______________
Chareyre Bruno
Maître de Conférences

Grenoble INP
Laboratoire 3SR - bureau E145
BP 53 - 38041, Grenoble cedex 9 - France
Tél : 33 4 56 52 86 21
Fax : 33 4 76 82 70 43
________________