# encoding: utf-8
"""
Overview
=======
This module contains breakage functions (bf) that can be used for particle breakage by replacement approach. Functions can be used for both spheres and clumps of spheres. However, this module is particularly useful for clumps because it deals with multiple clump-specific issues:
* Clump members do not interact. Hence, modification of the Love-Webber stress tensor is proposed to mimic interactions between clump members when the stress state is computed.
* If clumped spheres overlap, their total mass and volume are bigger than the mass and volume of the clump. Thus, clump should not split by simply releasing clump members. The mass of new sub-particles is adjusted to balance the mass of a nonoverlapping volume of the broken clump member.
* New sub-particles can be generated beyond the outline of the broken clump member to avoid excessive overlapping. Particles are generated taking into account the positions of neighbor particles and additional constraints (e.g. predicate can be prescribed to make sure that new particles are generated inside the container).
Clump breakage algorithm
=======
The typical workflow consists of the following steps (full description in [Brzezinski2022]_):
* Stress computation of each clump member. The stress is computed using the Love-Weber (LV) definition of the stress tensor. Then, a proposed correction of the stress tensor is applied.
* Based on the adopted strength criterion, the level of effort for each clump member is computed. Clump breaks if the level of effort for any member is greater than one. Only the most strained member can be split in one iteration.
* The most strained member of the clump is first released from the clump and erased from simulation. New mass and moment of inertia are computed for the new clump. The difference between the “old" and the “new" mass must be replaced by new bodies in the simulation.
* New, smaller spheres are added to the simulation balancing the mass difference. The spheres are placed in the void space, hence do not overlap with other bodies that are already present in the simulation (`splitting_clump`_).
* Finally, the soundness of the remaining part of the original clump needs to be verified. If the clump members do not contact each other anymore, the clump needs to be replaced with smaller clumps/spheres (`handling_new_clumps`_).
* Optionally, overlapping between new sub-particles of sub-particles and existing bodies can be allowed (`packing_parameters`_).
.. _splitting_clump:
.. figure:: fig/clump-breakage-splitting-clump.*
:scale: 35 %
:align: center
Stages of creating a clump in Yade software and splitting due to the proposed algorithm: (a) overlapping bodies, (b) clumped body (reduced mass and moments of inertia), (c) selection of clump member for splitting, (d) searching for potential positions of sub-particles, (e) replacing clump member with sub-particles, updating clump mass and moments of inertia.
.. _handling_new_clumps:
.. figure:: fig/clump-breakage-handling-new-clumps.*
:scale: 35 %
:align: center
Different scenarios of clump splitting: (a) clump remains in the simulation - only updated, (b) clump is split into spheres, (c) clump is split into a sphere and a new clump.
.. _packing_parameters:
.. figure:: fig/clump-breakage-packing-parameters.*
:scale: 35 %
:align: center
Replacing sphere with sub-particles: (a-c) non-overlapping, (d-f) overlapping sub-particles and potentially overlapping with neighbor bodies, (g-i) non-overlapping sub-particles but potentially overlapping with neighbor bodies.
Functions required for clump breakage algorithm described in:
Brzeziński, K., & Gladky, A. (2022), Clump breakage algorithm for DEM simulation of crushable aggregates. [Brzezinski2022]_
Strength Criterion adopted from;
Gladkyy, A., & Kuna, M. (2017). DEM simulation of polyhedral particle cracking using a combined Mohr–Coulomb–Weibull failure criterion. Granular Matter, 19(3), 41. [Gladky2017]_
:ysrc:`Source code file<py/bf.py>`
"""
from yade import pack
import numpy as np
# required when functions are put in an external module
from yade import Sphere, math
from yade.utils import sphere, growParticle
[docs]
def stressTensor(b, stress_correction=True):
"""
Modification of Love-Weber stress tensor, that applied to the clump members gives results similar to standalone bodies.
"""
center = b.state.pos
radius = b.shape.radius
volume = (np.pi * radius**3) * 4 / 3
sigma = np.zeros([3, 3]).astype(np.float32)
interactions = b.intrs()
if b.isStandalone: # if not clumped
f_u = O.forces.f(b.id) # sum off forces for stress correction
t_u = O.forces.t(b.id) # sum off forces for stress correction
for ii in interactions:
pp = ii.geom.contactPoint
contact_vector = pp - center
cont_normal = ii.geom.normal
ss = ii.phys.shearForce * np.dot(contact_vector / np.linalg.norm(contact_vector), -cont_normal)
nn = ii.phys.normalForce * np.dot(contact_vector / np.linalg.norm(contact_vector), -cont_normal)
ff = ss + nn
for i in range(3):
for j in range(3):
sigma[i, j] += ff[i] * contact_vector[j]
else:
c_id = b.clumpId
# sum off forces for stress correction
f_u = np.zeros(3).astype(np.float32)
# sum of torques for stress correction
t_u = np.zeros(3).astype(np.float32)
for ii in interactions:
# id of the other body (the one that b is contacting with)
the_other_id = list({ii.id1, ii.id2}.difference({b.id}))[0]
# ommit interaction if the contacting body belongs to the same clump
if O.bodies[the_other_id].clumpId == c_id:
continue
pp = ii.geom.contactPoint
contact_vector = pp - center
cont_normal = ii.geom.normal
ss = ii.phys.shearForce * np.dot(contact_vector / np.linalg.norm(contact_vector), -cont_normal)
nn = ii.phys.normalForce * np.dot(contact_vector / np.linalg.norm(contact_vector), -cont_normal)
ff = ss + nn
f_u += ff
t_u += np.cross(contact_vector, ss)
for i in range(3):
for j in range(3):
sigma[i, j] += ff[i] * contact_vector[j]
# at least unbalanced force [Brzeziński and Gladkyy 2022] magnitude need to be greater than need to be greater than zero
if stress_correction and np.linalg.norm(f_u) > 0:
f_u_prim = -1 * f_u
normal_prim = f_u / np.linalg.norm(f_u)
f_t_prim = np.cross(0.5 * t_u, normal_prim) / radius
f_t_bis = np.cross(0.5 * t_u, -1 * normal_prim) / radius
false_interactions = [
[radius * normal_prim, -1 * radius * normal_prim], # contact vectors in p_prim and p_bis respectively
[f_u_prim + f_t_prim, f_t_bis]
] # contact forces in p_prim and p_bis respectively
for ii in range(len(false_interactions)):
contact_vector = false_interactions[0][ii]
ff = false_interactions[1][ii]
for i in range(3):
for j in range(3):
sigma[i, j] += ff[i] * contact_vector[j]
sigma = sigma / volume
return sigma
[docs]
def checkFailure(b, tension_strength, compressive_strength, wei_V0=0.01, wei_P=0.63, wei_m=3, weibull=True, stress_correction=True):
"""
Strength criterion adopted from [Gladkyy and Kuna 2017]. Returns particles 'effort' (equivalent stress / strength) if the strength is exceeded, and zero othervise.
"""
sigma = stressTensor(b, stress_correction=stress_correction)
if weibull: # I just modify potential sigma_eff before checking conditions
V = (np.pi * b.shape.radius**3) * 4 / 3
coeff = ((-wei_V0 / V) * np.log(1 - wei_P))**(1 / wei_m)
sigma /= coeff # In gladky's paper there ia multiplication but sigma_eff_hat denotes strength, so I divide the acting stress
# The next line does not work with the typ mpf! That is why we are skipping the check tests with mpf
sigma_1, sigma_2, sigma_3 = np.sort(math.eig(sigma)[0])[::-1]
sigma_tau = sigma_1 - sigma_3 * tension_strength / compressive_strength
# case (a) # from [Gladkyy and Kuna 2017]
if sigma_1 < 0 and sigma_3 < 0 and sigma_3 < -compressive_strength:
sigma_eff = -sigma_3
effort = sigma_eff / compressive_strength
# case (b)
elif sigma_1 > 0 and sigma_3 > 0 and sigma_1 > tension_strength:
sigma_eff = sigma_1
effort = sigma_eff / tension_strength
# case (c)
elif abs(sigma_tau) > abs(tension_strength):
sigma_eff = abs(sigma_tau)
effort = sigma_eff / abs(tension_strength)
else:
effort = 0
return effort
[docs]
def replaceSphere(
sphere_id,
subparticles_mass=None,
radius_ratio=2,
relative_gap=0,
neighbours_ids=[],
initial_packing_scale=1.5,
max_scale=3,
scale_multiplier=None,
search_for_neighbours=True,
outer_predicate=None,
grow_radius=1.,
max_grow_radius=2.0
):
"""
This function is intended to replace sphere with subparticles.
It is dedicated for spheres replaced from clumps (but not only). Thus, two features are utilized:
- subparticles_mass (mass of the subparticles after replacement), since in a clump only a fraction of original spheres mass is taken into account
- neighbours_ids - list of ids of the neighbour bodies (e.g. other clump members, other bodies that sphere is contacting with) that we do not want to penetrate with new spheres (maybe it could be later use to avoid penetration of other bodies). However, passing neighbours_ids is not always necessary. By default (if search_for_neighbours==True), existing spheres are detected authomatically and added to neighbours_ids. Also, outer_predicate can beused to avoid penetrating other bodies with subparticles.
Spheres will be initially populated in a hexagonal packing (predicate with dimension of sphere diameter multiplied by initial_packing_scale ). Initial packing scale is greater than one to make sure that sufficient number of spheres will be produced to obtain reguired mass (taking into account porosity).
scale_multiplier - if a sufficient number of particles cannot be produced with initial packing scale, it is multiplied by scale multiplier. The procedure is repeated until initial_packing scale is reached. If scale_multiplier is None it will be changed to max_scale/initial_packing_scale, so the maximum range will be achieved in second iteration.
max_scale - limits the initial_packing_scale which can be increased by the algorithm. If initial_packing_scale >max_scale, sphere will not be replaced (broken).
outer_predicate - it is an additional constraint for subparticles generation. Can be used when non spherical bodies are in vicinity of the broken particle, particles are in box etc.
search_for_neighbours - if True searches for additional neighbours (spheres whithin a range of initial_packing_scale *sphere_radius)
Particles can be generated with smaller radius and than sligtly growed (by "grow_radius"). It allows for adding extra potential energy in the simulation, and increase the chances for successful packing.
relative_gap - is the gap between packed subparticles (relalive to the radius of subparticle), note that if grow_radius > 1, during subparticles arangemenr their radius is temporarily decreased by 1/grow radius. It can be used to create special cases for overlapping (described in the paper).
"""
assert grow_radius <= max_grow_radius
sphere = O.bodies[sphere_id]
sphere_radius = sphere.shape.radius
sphere_center = sphere.state.pos
full_mass = sphere.state.mass
if subparticles_mass == None:
subparticles_mass = full_mass
if subparticles_mass == 0:
print(
"Sphere (id={:d}) was erased without replacing with subparticles. The change of the mass is negligible (the updated mass of the clump has not changed). For grater accuracy increase the 'discretization' parameter."
.format(sphere_id)
)
O.bodies.erase(sphere_id)
return None
if scale_multiplier == None:
scale_multiplier = max_scale / initial_packing_scale
full_number_of_subparticles = radius_ratio**(3)
req_number_of_spheres = int(np.ceil(full_number_of_subparticles * subparticles_mass / full_mass))
subparticle_mass = subparticles_mass / req_number_of_spheres
sphere_mat = sphere.material
density = sphere_mat.density
exact_radius = ((3 * subparticle_mass) / (4 * np.pi * density))**(1 / 3)
# search for neighbours only once before the while loop
# add to the list neighbours (spheres) whithin the range of max_scale (maximum_initial_packing_scale radius)
if search_for_neighbours:
neighbour_set = set(neighbours_ids)
for bb in O.bodies:
if bb.id != sphere_id and isinstance(bb.shape, Sphere):
dist = ((np.array(sphere_center) - np.array(bb.state.pos))**2).sum()**0.5
if dist < sphere_radius * max_scale + bb.shape.radius:
neighbour_set.add(bb.id)
neighbours_ids = list(neighbour_set)
# populate space with new particles
no_gen_subparticles = 0
while no_gen_subparticles < req_number_of_spheres: # make sure enought subparticles were create
# make sure that there is enough of the particles
master_predicate = pack.inSphere(sphere_center, sphere_radius * initial_packing_scale)
for neighbour_id in neighbours_ids: # only works for spheres
nb = O.bodies[neighbour_id]
master_predicate = master_predicate - pack.inSphere(nb.state.pos, nb.shape.radius)
if outer_predicate != None:
master_predicate = master_predicate & outer_predicate
sp = pack.regularHexa(master_predicate, radius=exact_radius / grow_radius, gap=relative_gap * exact_radius, material=sphere_mat)
no_gen_subparticles = len(sp)
if no_gen_subparticles < req_number_of_spheres:
initial_packing_scale *= scale_multiplier
if initial_packing_scale > max_scale:
print("We will NOT crash the particle with id id={:d}. ).".format(sphere.id))
return sphere.id
# make sure that the number of the particles is exact, remove the furthest subparticles
sq_distances = []
for i in range(len(sp)):
sq_distance = (np.array(sp[i].state.pos - sphere_center)**2).sum()
# create a list that will store bodies and squared distances form the final master_sphere.
sq_distances += [[sp[i], sq_distance]]
# sort by distance - ascending
sq_dist_sorted = sorted(sq_distances, key=lambda x: x[1])
for i in range(req_number_of_spheres):
b_id = O.bodies.append(sq_dist_sorted[i][0])
growParticle(b_id, grow_radius)
# erase interactions and then sphere
ii = O.bodies[sphere_id].intrs()
for i in ii:
O.interactions.erase(i.id1, i.id2)
O.bodies.erase(sphere_id)
return None
[docs]
def evalClump(
clump_id,
radius_ratio,
tension_strength,
compressive_strength,
relative_gap=0,
wei_V0=0.001,
wei_P=0.63,
wei_m=3,
weibull=True,
stress_correction=True,
initial_packing_scale=1.5,
max_scale=3,
search_for_neighbours=True,
outer_predicate=None,
discretization=20,
grow_radius=1.,
max_grow_radius=2.0
):
"""
Iterates over clump members with "checkFailure" function.
Replaces the broken clump member with subparticles.
Split new clump if necessary.
If clump is not broken returns False, if broken True.
"""
members_ids = O.bodies[clump_id].shape.members.keys()
clump_mass = O.bodies[clump_id].state.mass
old_clump_pos = O.bodies[clump_id].state.pos
max_effort = 0
member_to_break = -1
for member_id in members_ids:
effort = checkFailure(
b=O.bodies[member_id],
tension_strength=tension_strength,
compressive_strength=compressive_strength,
wei_V0=wei_V0,
wei_P=wei_P,
wei_m=wei_m,
weibull=weibull,
stress_correction=stress_correction
)
if effort > max_effort:
max_effort = effort
if effort >= 1:
member_to_break = member_id
if max_effort < 1:
return False
# breaking algorithm for two-sphere clump
if len(members_ids) == 2: # if only two spheres in the clump
for member_id in members_ids:
if member_id != member_to_break:
conserved_mass = O.bodies[member_id].state.mass
subparticles_mass = clump_mass - conserved_mass
unbroken_sphere_id = replaceSphere(
member_to_break,
subparticles_mass=subparticles_mass,
radius_ratio=radius_ratio,
relative_gap=relative_gap,
initial_packing_scale=initial_packing_scale,
max_scale=max_scale,
search_for_neighbours=search_for_neighbours,
outer_predicate=outer_predicate,
max_grow_radius=max_grow_radius,
grow_radius=grow_radius
)
if not (unbroken_sphere_id is None): # Finish if sphere cannot be broken
return False
# If the other sphere was broken, erase clump (but leave the sphere that wasn't broken).
O.bodies.erase(clump_id)
# breaking algorithm for more than two-sphere clump
if len(members_ids) > 2:
O.bodies.releaseFromClump(member_to_break, clump_id, discretization=discretization)
conserved_mass = O.bodies[clump_id].state.mass
subparticles_mass = clump_mass - conserved_mass
# The line below is the same as in case of len(members_ids) == 2, but I can't do it outside condition because I need to check subparticles soundness in the same block of code
unbroken_sphere_id = replaceSphere(
member_to_break,
subparticles_mass=subparticles_mass,
radius_ratio=radius_ratio,
relative_gap=relative_gap,
initial_packing_scale=initial_packing_scale,
max_scale=max_scale,
search_for_neighbours=search_for_neighbours,
outer_predicate=outer_predicate,
max_grow_radius=max_grow_radius,
grow_radius=grow_radius
)
# if sphere cannot be broken first add it back to the clump, and then finish
if not (unbroken_sphere_id is None):
O.bodies.addToClump([member_to_break], clump_id, discretization=discretization)
return False
# check if all the subparticles in the clump are connected otherwise create separate clumps/spheres
list_of_sets = [] # sets to store members overlaping with the member 1 in each loop
new_member_ids = O.bodies[clump_id].shape.members.keys()
for member1 in new_member_ids:
list_of_sets.append({member1})
for member2 in new_member_ids:
r1 = O.bodies[member1].shape.radius
r2 = O.bodies[member2].shape.radius
p1 = np.array(O.bodies[member1].state.pos)
p2 = np.array(O.bodies[member2].state.pos)
dist = ((p1 - p2)**2).sum()**0.5
if dist < r1 + r2:
list_of_sets[-1].add(member2)
# Now sets are prepared, so verify whether there are any separete clumps.
# The sets represent all the bodies that every member contacts with.
# Union the intersecting sets until only distinct sets of elements remain.
union_count = 1
while union_count > 0:
union_count = 0
for i in range(len(list_of_sets) - 1):
for j in range(i + 1, len(list_of_sets)):
s1 = list_of_sets[i]
s2 = list_of_sets[j]
if len(s1.intersection(s2)) > 0:
list_of_sets[i] = s1.union(s2)
list_of_sets[j].clear()
union_count += 1
else:
continue
# This way I should have empty not important sets, sets of single values to replace by spheres, sets of multiple values to replace with clumps.
# If any set before last one will not be empty, the whole clump needs to be erased. That means members will be erased to constitute other clumps and/or standalone spheres.
len_of_list = len(list_of_sets)
erase_clump = False
for i in range(len(list_of_sets)):
if len(list_of_sets[i]) == 0:
continue
else:
if i < len_of_list:
erase_clump = True
if len(list_of_sets[i]) == 1:
member_id = list(list_of_sets[i])[0]
O.bodies.append(sphere(O.bodies[member_id].state.pos, O.bodies[member_id].shape.radius, material=O.bodies[member_id].material))
O.bodies.erase(member_id)
if len(list_of_sets[i]) > 1:
members_ids = list(list_of_sets[i])
list_of_spheres = []
for member_id in members_ids:
list_of_spheres += [
sphere(O.bodies[member_id].state.pos, O.bodies[member_id].shape.radius, material=O.bodies[member_id].material)
]
O.bodies.erase(member_id)
O.bodies.appendClumped(list_of_spheres, discretization=discretization)
if erase_clump:
O.bodies.erase(clump_id)
return True
