particle blow up

Asked by ytang on 2020-11-27

Hi all,

I'm simulating the CPT tests. I use the geom.facetCylinder and geom.facetCone commands to generate the penetrator.
the code shows below:
#############################################################
x0=0.2;y0=0.2;z0=0.39665
cylinderIDS= O.bodies.append(geom.facetCylinder((x0,y0,0.35665),
 radius=0.0125,height=0.15,orientation=Quaternion((1, 0, 0), 0),wallMask=5,segmentsNumber=10, angleRange=None,material=facetMat))
coneIDS= O.bodies.append(geom.facetCone((x0,y0,0.270825),
 radiusTop=0.0125,radiusBottom=0.0,height=0.02165,orientation=Quaternion((1, 0, 0), 0),wallMask=6,segmentsNumber=10, angleRange=None,material=facetMat))
#######################################
I did two tests. one is 10 segments number, another one is 50 segments number for the cylinder and cone.
###############################
for the 10 segments case: the simulation runs without particles blowing up. while for the 50 segments case, the particles blowing up after the penetrator touches the sample surface.
#########################################
These two simulations are exactly the same except for the segment number. for these two cases, I set the same timestep.
#######################################################
GlobalStiffnessTimeStepper(active=1,timeStepUpdateInterval=10,timestepSafetyCoefficient=0.8),
###################
###################

I think the particles blowing up may result from the time step.

Do you have any advice?
Thanks!

Question information

Language:
English Edit question
Status:
Solved
For:
Yade Edit question
Assignee:
No assignee Edit question
Solved by:
Jan Stránský
Solved:
2020-11-29
Last query:
2020-11-29
Last reply:
2020-11-28
Jan Stránský (honzik) said : #1

Hello,

> I think the particles blowing up may result from the time step.

looks like this.
If a sphere interacts with multiple facets, the add force to it independently.
In the case of edges, the resulting repulsive force is two times what should be (sum force from two facets).
Even worse are vertices, the worst is the tip of cylinder, where there are not two but 50 facets and the force is 50 time what should be.
Essentially it is like the cone tip has 50x higher stiffness, resulting in the "time step problem" not considering this phenomenon.

Consider using PFacets instead.

You can try to debug the problem (e.g. printing forces on the "tip" sphere and comparing with expectation, trying (much) lower timestepSafetyCoefficient etc.)

cheers
Jan

ytang (ytang116) said : #2

Hi Jan,

here is my understanding of your answer. if it isn't right, please correct me.
##########################################################
If a sphere contact with two facets in the edge, the repulsive force will be two times.
For example, if the sphere contact with one facet (not in the edge), the repulsive force is 1N. The repulsive force for the sphere contact with two facets in the edge will be 2N (if we consider the sphere-facet contact force is the same for these two cases).
################################################
If I set 1000 segments for the cone, the repulsive force at the tip of the cone will be 100 times larger than that of the cone with 10 segments. (maybe not 100 times, but the repulsive force will be much higher than the cone with 10 segments)
################################################
################################################
If this is the case, how can we get the real penetration resistance when the penetrator penetrates into the sample?
I mean, different segment number has different penetration resistance.
################################################
> You can try to debug the problem (e.g. printing forces on the "tip" sphere and comparing with expectation, trying (much) lower timestepSafetyCoefficient etc.)

As for the time step, I will try to decrease the parameter.
###############################################
> Consider using PFacets instead.
As for the PFacets, does this will solve the sphere-edges/sphere-vertices contact problem?
or what is the effect of using PFacets? can you explain more about this?

best,
ytang

Best Jan Stránský (honzik) said : #3

> how can we get the real penetration resistance when the penetrator penetrates into the sample?

using PFacets

> As for the PFacets, does this will solve the sphere-edges/sphere-vertices contact problem?
> or what is the effect of using PFacets? can you explain more about this?

The main purpose of PFacets (IMO) is to improve this facet behavior.
PFacets consist of vertex/node - edge - facet hierarchy.
If the contact od of sphere-vertex type, the interaction is considered with this one vertex, not any more with more vertices or edges of facets.
If the contact is of sphere-edge type, the interaction is considered with this one edge, not any more with more facets.
Finally the sphere-face type of contact is handled naturally.

So for this problem, pfacets are worth to try at least.

cheers
Jan

ytang (ytang116) said : #4

Thanks Jan Stránský, that solved my question.