yade-users team mailing list archive

[Bruno Chareyre <bruno.chareyre@xxxxxxxxxxx>]

For those interested, I elaborated the comment a little in ScGeom.cpp 
(possible wiki paragraph in the future), as this "granular ratchetting" 
needed explanation.
We could put a simple py script to simulate the cycle explained below, 
and test any law in Yade to see if it generates ratchetting.

Cheers

Bruno


"The following definition of c1x and c2x is to avoid "granular 
ratcheting". This phenomenon has been introduced to me by S. McNamara in 
a seminar help in Paris, 2004 (GDR MiDi). The concept is also mentionned 
in many McNamara, Hermann, Lüding, and co-workers papers (see online : 
"Discrete element methods for the micro-mechanical investigation of 
granular ratcheting", R. García-Rojo, S. McNamara, H. J. Herrmann, 
Proceedings ECCOMAS 2004, @ 
http://www.ica1.uni-stuttgart.de/publications/2004/GMH04/), where it 
refers to an accumulation of strain in cyclic loadings.
        
       Unfortunately, published papers tend to focus on the "good" 
ratcheting, i.e. irreversible deformations due to the intrinsic nature 
of frictional granular materials, while the talk of McNamara in Paris 
clearly mentioned a possible "bad" ratcheting, purely linked with the 
formulation of the contact laws in what he called "molecular dynamics" 
(i.e. Cundall's model, as opposed to "contact dynamics" from Moreau and 
Jean).
      
       Giving a short explanation :
       The bad ratcheting is best understood considering this small 
elastic cycle at a contact between two grains : assuming b1 is fixed, 
impose this displacement to b2 :
       1. translation "dx" in the normal direction
       2. rotation "a"
       3. translation "-dx" (back to initial position)
       4. rotation "-a" (back to initial orientation)
      
       If the branch vector used to define the relative shear in 
rotation*branch is not constant (typically if it is defined from the 
vector center->contactPoint like in the "else" below), then the shear 
displacement at the end of this cycle is not null : rotations a and -a 
are multiplied by branches of different lengths.
       It results in a finite contact force at the end of the cycle 
even though the positions and orientations are unchanged, in total 
contradiction with the elastic nature of the problem. It could also be 
seen as an inconsistent energy creation or loss. It is BAD! And given 
the fact that DEM simulations tend to generate oscillations around 
equilibrium (damped mass-spring), it can have a significant impact on 
the evolution of the packings, resulting for instance in slow creep in 
iterations under constant load.
       The solution to avoid that is quite simple : use a constant 
branch vector, like here radius_i*normal."

--
_______________
Bruno Chareyre
Associate Professor
Grenoble INP
Lab. 3SR
BP 53 - 38041, Grenoble cedex 9 - France
Tél : 33 4 56 52 86 21
Fax : 33 4 76 82 70 43
________________