Hi, not sure this is related to logarithmic strain -- it is called dynamic/adaptive timestepping and is implemented in the woo.dem.DynDt engine (also see Timestep computation (programming manual)). This engine is automatically added when using woo.dem.DemField.minimalEngines (which actually is woo.utils.defaultEngines) as is nost preprocessors and examples. Using the wrapper ensures that base engines are in the correct order (integrator, collider, contacts, dynamic timestepper) but you can do just the same by hand as well. Actually, there is very rarely the need to set S.dt by hand as DynDt and the machinery around should be deal with usual simulations. HTH, Vaclav
sorry I mixed up 2 different things. Have you implemented logarithmic strain in your code yet? I have seen somewhere you mentioned about this strain formulation.
Are we talking about periodic boundary conditions? The documentation being somewhat incomplete, I refer you to https://github.com/woodem/woo/blob/master/core/Cell.hpp#L7 which says that Cell::trsf is deformation gradient tensor (I am updating the documentation about that right now). That should be a base to compute any other strain measure you may need. To be honest, I am not very good with finite strain theories, so double-check. HTH, vaclav
Ps if you are referring to https://yade-dem.org/doc/formulation.html#variables, that was written by myself, the logarithmic strain was used for the concrete model to avoid interpenetration under high confinement. IIRC it never worked really in 3D as the ratio between shear and normal forces changes (quite a bit) and the interpenetrating sphere would slide sideways and get through anyway. The current concrete model Woo does not have that, looking at the equation phys.epsN=(geom.uN-phys.uN0)/geom.lens.sum() here. v.
thank you for your helpful information, that clears all my doubts.
Yes I was referring to that strain formulation, and I also ran uniaxial compression simulations on concrete particles with confinement, which yield strange results and particles was flying away) in the unconfined direction, I also checked the ConcretePhys and figured that out... Would you suggest to run these simulations with periodic boundary condition?
Could you explain more about the changes between shear and normal force that cause particles to penetrate each others?
all in all thanks for such quick support.
The interpenetration comes from the fact that DEM is designed for relatively small overlaps which represent deformations; this is achieved via various stiffness parameters. Higher compression (we talk 1D here) would in reality lead to highly non-linear behavior, with strong repulsive forces, but in DEM, this is beyond the usual scope and thus the model fails to reproduce that well. It could be theoretically perhaps handles (like in the Hertz models, but those are also derived for small deformations) but increased stiffness leads to strain-dependent critical timestep, meaning a small process zone with high strain leads to small timestep for the whole simulation; this is what I wanted to avoid by using constant-stiffness contacts. The logarithmic strain is a workaround, but it actually would also increase stiffness (thus timestep) and secondly, while it look okay-ish in 1D (linear strain), it has numerous issues in 3D. In 3D, the particle can easily slip sideways, which is normally prevented by shear stiffness, but shear stiffness does not increase with logarithmic strain, so as the two are getting closed, shear resistance gets (in proportion) smaller. Another issue is that the contact behavior is not conservative anymore (there can be spurious energy in cyclic loading, reasulting from interplay between normal force and shear dissipation) which really means that it is physically nonsensical.
Another reason for exploding particles could be too high a strain rate. Check that you are below something like 1e-3/s (IIRC), otherwise inertia effects will be strong.
Does it make sense? Sorry for somewhat superficial answer just from the top of my head. I forgot a lot already :))
I understand you points... Now I am getting confused with the model. It seems that in 3D it is harder to predict particle behavior and every model and method have its own scope and limitation...
During the study of the concrete model in DEM, at first I did not understand your confinement parameters (hence I turned it off for most simulations). But it seems to somewhat make the interpenetrating issue become less (as in one of your paper mentioned that this approach somehow take care of boundary confinement at contact level). Is there an intuition behind this approach? for the moment, in my simulations, I just set the particles on the boundary to only be able to move vertically (unconfined direction), I guess it is then not equivalent of sufficient?
Thank again, Vaclav
Hi, you get the point, DEM models are like that. It stems partially from the fact that DEM is not formulated based on continuum theory, it is instead closer to lattices or agent systems; thus there no rigorous notions of exact solution, convergence etc like in FEM, which can guide you. The job is to predict continuum (macroscopic) behavior from contact-level behavior, sometimes referred to as micro-macro transition. Some DEM models use continuum-like measures (like this concrete model used at some point volumetric strain around particle) to get more information in. Given its origins and ideas behind it, using DEM for continuum-like (not particulate) materials will always be a little bit of hack around its (perhaps too simple, for those tasks) principles.
The confinement was to simulate confined environment (it was concrete under confinement from the boundary) by adding the confinement to the contact law directly, without using DEM for it; so actually the DEM only simulated the difference from the confined state, something like that. But I don't remember exactly, to be honest.