Hi Ruochun, Thanks for your insight! I suspect that the difference in contact models is likely the discrepancy between my work and the other paper. I'll see if I can implement the contact model that they had used and if it will improve my results. I'm currently working to make a model that aligns better with another experiment to perform a more proper validation exercise.
Thanks, David On Wednesday, July 13, 2022 at 2:23:23 PM UTC-6 Ruochun Zhang wrote: > Hi David, > > This is a very involved question because, I imagine that only people with > research experience of pebble flow problems can confidently step forward > and claim to offer "technical insights". I have never researched said > problems. There are a number of suggestions however, that Luning and I > think we can offer. > > First please make sure the scale/resolution of the simulations matches. > The number and grain size of particles can make a big difference. It seems > the data in this paper is measured under certain equilibrium conditions, > that is also something to pay attention to. > > Then perhaps more importantly, I am pretty sure these 2 implementations do > not use the exact same contact model. The normal damping term seems to be > slightly different and, it seems this paper did not mention how the > tangential stiffness and damping coefficients should be calculated. This > could be key information: in this quasi-static problem, the tangential > force is directly in the "downward" flow direction and should play an > essential role. This is true even when you have the same friction > coefficient 0.3, because 1) the normal force may be different; 2) I imagine > for many contact pairs this tangential force is in the "micro displacement" > region and not equal to 0.3 times normal force. Also, I didn't quite follow > the explanation for the differences between the Hookean and Hertzian models > in the paper, that supposedly caused the discrepancy we see in Fig. 9. How > are they different in the paper? In short, it's quite likely we are not > using the same force models, even though they are both Hertzian model; but > just from this paper we cannot verify that. > > Following the previous argument, from a quick read-through I don't know if > it's indeed because of the discrepancy between Hookean and Hertzian, as > opposed to the discrepancy between the exact models Rycroft and Sun et al > used. Again, maybe it is well-known in that research community, it's just > that I don't know. My engineering intuition though, is that this should be > similar to a CFD boundary layer problem, where you see sharp near-boundary > velocity decline for non-viscos flows, and "flatter" velocity profiles for > viscos flows. And following this line of thinking, you may be able to match > the velocity profile that you are after, by correctly modelling the flow > "viscosity". Maybe you can try adding cohesion (cohesion coefficient of > something around 10) and/or some rolling resistance and/or modified > tangential stiffness, to see if it achieves something similar. You probably > would not use this "empirical solution" in the end, but it can be a sanity > check, so we know whether some "subtle" settings can be useful in capturing > this physics, even as useful as using a particular class of model. > > I noticed many DEM users call for "slightly customized" force models, that > is something I try to reflect on the APIs of the upcoming next-gen Chrono > DEM solver. But with current Chrono::GPU, you have to go to function > *materialPropertyCombine > *to modify the material-property-to-normal-tangential-stiffness > conversion, and to go to function *computeSphereContactForces_matBased *to > modify contact force calculations, then rebuild the project. You could do > that and maybe we can help you do that, if a clear definition of the force > model is given, and you would like to reproduce that result. > > I cannot make comments specific to the pebble flow problems, and someone > with the research experience can definitely do a better job; but I hope > this helps. > > Thank you, > Ruochun > On Wednesday, July 13, 2022 at 10:36:54 AM UTC-5 [email protected] wrote: > >> Hello, >> >> I have been running some granular flow simulations with the GPU module. >> When comparing my results to other similar work that uses other codes, my >> results are not aligning with those from the other works, even when the >> same material properties are used. I cannot post my results publicly on the >> forum, but can share them with someone from the Chrono team over email. >> >> Essentially, I am seeing that in previous works, the streamwise velocity >> of particles is continuously decreasing from the center of the granular bed >> towards the wall. In my simulation, however, the velocity of particles is >> essentially the same across the bed, except for the particles that directly >> touch the wall. >> >> The paper I am comparing to is: >> X. Sun, Three-Dimensional numerical simulation of quasi-static pebble flow >> >> In particular from this paper in Figure 9, my results look much more like >> the Hookean velocity profiles than the Hertzian profiles that agree with >> experimental results. Any help on understanding what is going on would be >> appreciated. >> >> Thanks! >> David >> > -- You received this message because you are subscribed to the Google Groups "ProjectChrono" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/projectchrono/ba694786-2596-4274-bb62-60781a2e8a2dn%40googlegroups.com.
