Rolling resistance - Problem in 3D?

Submitted by THomas Pettersson on Thu, 02/26/2015 - 10:21

Hello!

I'm looking into the subject of rolling resistance and have been reading the article "Assessment of rolling resistance models in discrete element simulations" by Jun Ai, Jian-Fei Chen, J. Michael Rotter and Jin Y. Ooi, as suggested by xinze in an older forum post (http://www.cfdem.com/forums/references-rolling-firction-dem). They claim that "the general conclusions are also applicable to 3D systems" but I find it hard to see how the suggested method is generalized to 3D, in which direction should the torque be directed? Doesn't summing up the vectors risk to direct the torque of a particle rolling in an arc (for example on a inclined plane) in some weird direction towards the starting point instead of backwards, and to force it in a backwards direction return the behaviour of the A type model. I also have the same question about similar models for tangential friction.

I also wonder about the fact that they doesn't limit the viscous part of the resistance until they reach maximum in the elastic part. I guess that that could lead to a considerable discontinuity in force amplitude and direction when the limitation kicks in. Wouldn't it be smother to add the forces together and then limit the sum?

How does LIGGGTHS usually handle thees issues and have anyone experienced problems related to them?

/Thomas

Forums:

JoshuaP | Thu, 02/26/2015 - 12:51

Hi,
I'm not sure but if you take a look in the rolling_model_epsd2_h_, the rolling resistance seems to be calculated for each direction and I cant see any problem in that. If you reach the maximum resistance moment it starts to roll but the maximum resistance moment is still acting on the particle, so that if the driving torque gets reduced it will stop rolling and remain in the elastic region.
Sure this moment isn't able to represent the real behavior of all grains but if you regard the macro-scale behavior you can obtain supporting structures which can represent real soil behavior well.

regards

JoshuaP | Thu, 02/26/2015 - 12:59

btw the maximum torque is always compared to the sum of all rollingmoments:
// limit max. torque
const double r_torque_mag = vectorMag3D(r_torque);
const double r_torque_max = fabs(cdata.Fn)*reff*rmu;
if(r_torque_mag > r_torque_max)

THomas Pettersson | Fri, 02/27/2015 - 09:10

Thanks for your quick answer JoshuaP!

Well you are right in that my second question isn't a problem in the epsd2 model, since the model have no viscous (no d if I haven't misunderstood the abbreviation) part. Does anyone use the first epsd model or is it considered obsolete?

regards

JoshuaP | Fri, 02/27/2015 - 10:54

hey,
you are welcome. I'm also using these models for my simulations and I'm interesting in every problems and questions about it.
For the limiting of the damping part, I found in the code that there is no damping when the critical torque is reached and it begins to roll. So the dashpot is just acting during the elastic part.

ckloss's picture

ckloss | Mon, 03/02/2015 - 09:46

Hi,

>> Does anyone use the first epsd model or is it considered obsolete?
No it is not obsolete, but EPSD2 is more convenient to use since it does not require to specify the damping part explicity (but is DOES have a damping part)

>>but I find it hard to see how the suggested method is generalized to 3D, in which direction should the torque be directed
I am not sure if I fully understood your line of argumentation. The torque is opposed to the relative rotational motion. in my understanding that works perfectly in 2D as well as in 3D

Best wishes
Christoph

THomas Pettersson | Tue, 03/24/2015 - 17:02

Hello Christoph!

Imagine that you are rolling a particle on a inclined plane with a rolling resistance big enough to stop the particle from rolling freely off the plane . If you roll it straight up you can recreate the result from the article mentioned above since we don't actually use the extra dimension. But now imagine that you roll it at an angle, lets say alpha. the angle will of course increasing since that it will be affected by gravity. In what direction is the torque now directed?

We can try to sum it up every time step, then we get an final direction that initially points in a straight line towards the start, (not along the path actually used to get there) and after max resistance is reached it will end up somewhere in between the starting point and the direction it comes from now. This could potentially be a big deviation in direction from opposed the relative rotational motion. Not to mention what will happen if the surface isn't a plane or if the particle gets hit by another particle or bounces against another plane while rolling.

So if we instead just sum the magnitude of the rolling resistance and put it opposed the relative rotational motion. Then we will have a kind of nice behavior that looks like one would expect until we arrive at the point where the particle is supposed to stop and we will return to what a situation that is very similar to the model that if I recall correctly is called Model A in the article, meaning that as soon as the particle stops it will for the computer reach a very small velocity facing the other direction, since it's virtually impossible to end at exactly 0 for float values. and it will be puched down our plane again. effectively not stopping but rolling in a kind of slow.

So my question was if there was a solution to the problem or if you just have to live with it.

ckloss's picture

ckloss | Fri, 04/03/2015 - 10:35

Hi Thomas,

>>Imagine that you are rolling a particle on a inclined plane with a rolling resistance big enough to stop the particle from rolling freely off the plane .
>> If you roll it straight up you can recreate the result from >>the article mentioned above since we don't actually use the extra dimension.
>>But now imagine that you roll it at an angle, lets say alpha. the angle will of course increasing since that it will be affected by
>>gravity. In what direction is the torque now directed?
Hi Thomas, I'm afraid I can't fully follow your line of argumentation.

But I agree, all of the rolling friction models in LIGGGHTS(R) (cdt,epsd,epsd2) are simplified, and will fail to picture the trajectory of a single particle correctly. But so do all other models if you look into every detail. For me as a modeller it's important that all of the rolling friction models can controll the angle of repose, and epsd and epsd2 can additionally depict a stable, unbounded heap

I am happy to continue discussion on the model details - I'm sorry I just could not follow your description above.

Best wishes
Christoph

JoshuaP | Tue, 04/28/2015 - 11:16

Hey,
since the rolling resistance Moment is depending on the radii of the contacting particles, is it necessary to use the same scaling of particle size in all simulations?
Considering a particle on a plane and we wanna try to push it (lets consider with a moving wall), it will occur different behaviour depending on the particle size.

regards
Joshua