Managing Clustered Networks#
This tutorial will cover the common issue of a network that is “clustered”, meaning that some pores are not connected the main body of the network. This is fairly common when extracting networks from tomogram, but can occur when getting creative with networks like stitching two together. A clustered network causes problems for the numerical solvers because it leads to “singular” matrices (i.e. they cannot be solved). Below we will explain what this means, and how to fix it.
import openpnm as op
import numpy as np
import matplotlib.pyplot as plt
pn = op.network.Demo(shape=[5, 5 ,1])
ax = op.visualization.plot_tutorial(pn, font_size=8)
Creating a Clustered Network#
Now that we’ve created a network, lets trim some throats to create clusters of isolated pores:
op.topotools.trim(network=pn, throats=[4, 5, 8, 9, 20, 21, 30, 31])
ax = op.visualization.plot_tutorial(pn, font_size=8)
Now we can have 3 total “clusters”, two small clusters at the bottom plus one large cluster which constitutes the rest of the network.
Illustrating the Problem#
Let’s first see the problem this creates and why. This requires running (or attempting to run) a transport simulation.
w = op.phase.Water(network=pn)
w.add_model(propname='throat.hydraulic_conductance',
model=op.models.physics.hydraulic_conductance.generic_hydraulic)
flow = op.algorithms.StokesFlow(network=pn, phase=w)
flow.set_value_BC(pores=pn.pores('left'), values=200_000)
flow.set_value_BC(pores=pn.pores('right'), values=100_000)
try:
flow.run()
except Exception as e:
print(e)
Your network is clustered, making Ax = b ill-conditioned
As we can see above, attempting to run flow resulted in an Exception with a message about the matrix being singular. Let’s see how a clustered network creates this problem. Let’s print the A matrix as an image:
fix, ax = plt.subplots(figsize=[5, 5])
A = np.array(flow.A.todense() != 0)
ax.pcolormesh(A, edgecolors='k', linewidth=0.5);
Referring to the network visualization above we can see that pores 5 and 10 are connected to each other, but not to any other pores. Looking at the A matrix above we can this reflected in the pattern. We can see that on row 5 there is an entry in the diagonal (5, 5) and a second entry at column 10 (5, 10). Now looking at row 10 we see an entry on the diagonal (10, 10) and a entry at column 5 (10, 5). These off-diagonal entries represent the conductance values between pores 5 and 10 (and between 10 and 5). The key point to note is that NO OTHER entries are present on either row. This means that the mass balance equations for pores 5 and 10 are not coupled to the rest of the equations. Another way to say is that pores 5 and 10 form their own independent matrix which can be solved in isolation since they don’t exchange mass with any of the other pores…however, we can’t solve this ‘mini-system’ because we have not specified any boundary conditions for it.
Fixing Clustered Networks#
This is easy to see by eye, but we need to tell OpenPNM to find them so we can begin to fix the problem. In version 3 we added many new pore-scale models to the models.network libary, among them are cluster_number and cluster_size. Let’s add these two models to the network and see what we get:
pn.add_model(propname='pore.cluster_number',
model=op.models.network.cluster_number)
pn.add_model(propname='pore.cluster_size',
model=op.models.network.cluster_size)
print(pn)
══════════════════════════════════════════════════════════════════════════════
net : <openpnm.network.Demo at 0x7fb3892ddda0>
――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――
# Properties Valid Values
――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――
2 pore.coords 25 / 25
3 throat.conns 32 / 32
4 pore.coordination_number 25 / 25
5 pore.max_size 25 / 25
6 throat.spacing 32 / 32
7 pore.seed 25 / 25
8 pore.diameter 25 / 25
9 throat.max_size 32 / 32
10 throat.diameter 32 / 32
11 throat.cross_sectional_area 32 / 32
12 throat.hydraulic_size_factors 32 / 32
13 throat.diffusive_size_factors 32 / 32
14 throat.lens_volume 32 / 32
15 throat.length 32 / 32
16 throat.total_volume 32 / 32
17 throat.volume 32 / 32
18 pore.volume 25 / 25
19 pore.cluster_number 25 / 25
20 pore.cluster_size 25 / 25
――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――
# Labels Assigned Locations
――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――
2 pore.xmin 5
3 pore.xmax 5
4 pore.ymin 5
5 pore.ymax 5
6 pore.surface 16
7 throat.surface 14
8 pore.left 5
9 pore.right 5
10 pore.front 5
11 pore.back 5
――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――――
We can see that cluster number and size have been added to the list of properties. Let’s inspect them:
pn['pore.cluster_number']
array([0., 0., 0., 0., 0., 1., 2., 0., 0., 0., 1., 2., 0., 0., 0., 0., 0.,
0., 0., 0., 0., 0., 0., 0., 0.])
We can see that most pores a labeled 0, meaning they belong to cluster 0. However, there are a few with cluster numbers of 1 and 2, which we know correspond to the two small clusters in our network. We could solve our current problem by trimming these pores, but let’s do it the proper way. Had we not looked at the network visualization, We would not actually know that cluster 0 is the main cluster. We can look at the size of each cluster instead:
pn['pore.cluster_size']
array([21., 21., 21., 21., 21., 2., 2., 21., 21., 21., 2., 2., 21.,
21., 21., 21., 21., 21., 21., 21., 21., 21., 21., 21., 21.])
Now we can see that most pores belong to a large cluster of size 21, where there are a few on small clusters. Now we can clearly tell which pores should be trimmed:
Ps = pn['pore.cluster_size'] < 21
op.topotools.trim(network=pn, pores=Ps)
ax = op.visualization.plot_tutorial(pn, font_size=8)
Now we can see only a single cluster. Without even re-running the pore-scale models we can see that the pores with a cluster number other than 0 have been removed:
pn['pore.cluster_number']
array([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
0., 0., 0., 0.])
Now we can run the flow algorithm without a problem, though note that we need to create a new algorithm since the previous one was created based on the number of pores and throats in the clustered network.
flow = op.algorithms.StokesFlow(network=pn, phase=w)
flow.set_value_BC(pores=pn.pores('left'), values=200_000)
flow.set_value_BC(pores=pn.pores('right'), values=100_000)
flow.run()
print(flow.x)
[200000. 200000. 200000. 200000.
200000. 185681.0100741 191927.4090955 179205.17687051
172109.57021197 168967.91198143 146211.55579289 113444.30802803
123898.09033738 162910.75854863 138766.19237225 124107.41400846
100000. 100000. 100000. 100000.
100000. ]
ax = op.visualization.plot_connections(network=pn, color_by=flow.rate(throats=pn.Ts, mode='single'))
op.visualization.plot_coordinates(network=pn, color_by=flow.x, ax=ax, markersize=100)
/home/runner/work/OpenPNM/OpenPNM/openpnm/visualization/_plottools.py:352: RuntimeWarning: overflow encountered in cast
color_by = np.array(color_by, dtype=np.float16)
[00:08:32] WARNING nans or infs found in color_by array, setting to 0 _plottools.py:357
<matplotlib.collections.PathCollection at 0x7fb32c5575e0>
