1
  2
  3
  4
  5
  6
  7
  8
  9
 10
 11
 12
 13
 14
 15
 16
 17
 18
 19
 20
 21
 22
 23
 24
 25
 26
 27
 28
 29
 30
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
import numpy as np

from py_diff_stokes_flow.env.env_base import EnvBase
from py_diff_stokes_flow.common.common import ndarray

class FlowAveragerEnv3d(EnvBase):
    def __init__(self, seed, folder):
        np.random.seed(seed)

        cell_nums = (64, 64, 4)
        E = 100
        nu = 0.499
        vol_tol = 1e-2
        edge_sample_num = 2
        EnvBase.__init__(self, cell_nums, E, nu, vol_tol, edge_sample_num, folder)

        # Initialize the parametric shapes.
        self._parametric_shape_info = [ ('bezier', 11), ('bezier', 11), ('bezier', 11), ('bezier', 11) ]
        # Initialize the node conditions.
        self._node_boundary_info = []

        inlet_range = ndarray([
            [0.1, 0.4],
            [0.6, 0.9],
        ])
        outlet_range = ndarray([
            [0.2, 0.4],
            [0.6, 0.8]
        ])
        cx, cy, _ = self.cell_nums()
        nx, ny, nz = self.node_nums()
        inlet_bd = inlet_range * cy
        outlet_bd = outlet_range * cy
        for j in range(ny):
            for k in range(nz):
                # Set the inlet at i = 0.
                if inlet_bd[0, 0] < j < inlet_bd[0, 1]:
                    self._node_boundary_info.append(((0, j, k, 0), 1))
                    self._node_boundary_info.append(((0, j, k, 1), 0))
                    self._node_boundary_info.append(((0, j, k, 2), 0))
                if inlet_bd[1, 0] < j < inlet_bd[1, 1]:
                    self._node_boundary_info.append(((0, j, k, 0), 0))
                    self._node_boundary_info.append(((0, j, k, 1), 0))
                    self._node_boundary_info.append(((0, j, k, 2), 0))
        # Set the top and bottom plane.
        for i in range(nx):
            for j in range(ny):
                for k in [0, nz - 1]:
                    self._node_boundary_info.append(((i, j, k, 2), 0))
        # Initialize the interface.
        self._interface_boundary_type = 'free-slip'

        # Other data members.
        self._inlet_range = inlet_range
        self._outlet_range = outlet_range
        self._inlet_bd = inlet_bd
        self._outlet_bd = outlet_bd

    def _variables_to_shape_params(self, x):
        x = ndarray(x).copy().ravel()
        assert x.size == 8

        cx, cy, _ = self._cell_nums
        lower = ndarray([
            [1, self._outlet_range[0, 0]],
            x[2:4],
            x[:2],
            [0, self._inlet_range[0, 0]],
        ])
        right = ndarray([
            [1, self._outlet_range[1, 0]],
            [x[4], 1 - x[5]],
            x[4:6],
            [1, self._outlet_range[0, 1]],
        ])
        upper = ndarray([
            [0, self._inlet_range[1, 1]],
            [x[0], 1 - x[1]],
            [x[2], 1 - x[3]],
            [1, self._outlet_range[1, 1]],
        ])
        left = ndarray([
            [0, self._inlet_range[0, 1]],
            x[6:8],
            [x[6], 1 - x[7]],
            [0, self._inlet_range[1, 0]],
        ])
        cxy = ndarray([cx, cy])
        lower *= cxy
        right *= cxy
        upper *= cxy
        left *= cxy
        params = np.concatenate([lower.ravel(),
            [0, -0.01, 1],
            right.ravel(),
            [0.01, 0, 1],
            upper.ravel(),
            [0, 0.01, 1],
            left.ravel(),
            [-0.01, 0, 1]
        ])

        # Jacobian.
        J = np.zeros((params.size, x.size))
        J[2, 2] = J[3, 3] = 1
        J[4, 0] = J[5, 1] = 1

        J[13, 4] = 1
        J[14, 5] = -1
        J[15, 4] = J[16, 5] = 1

        J[24, 0] = 1
        J[25, 1] = -1
        J[26, 2] = 1
        J[27, 3] = -1

        J[35, 6] = J[36, 7] = 1
        J[37, 6] = 1
        J[38, 7] = -1
        J[:, ::2] *= cx
        J[:, 1::2] *= cy
        return ndarray(params).copy(), ndarray(J).copy()

    def _loss_and_grad_on_velocity_field(self, u):
        u_field = self.reshape_velocity_field(u)
        grad = np.zeros(u_field.shape)
        nx, ny, nz = self.node_nums()
        loss = 0
        cnt = 0
        for j in range(ny):
            for k in range(nz):
                if self._outlet_bd[0, 0] < j < self._outlet_bd[0, 1] or \
                    self._outlet_bd[1, 0] < j < self._outlet_bd[1, 1]:
                    cnt += 1
                    u_diff = u_field[nx - 1, j, k] - ndarray([0.5, 0, 0])
                    loss += u_diff.dot(u_diff)
                    grad[nx - 1, j, k] += 2 * u_diff
        loss /= cnt
        grad /= cnt
        return loss, ndarray(grad).ravel()

    def _color_velocity(self, u):
        return float(np.linalg.norm(u) / 3)

    def sample(self):
        return np.random.uniform(low=self.lower_bound(), high=self.upper_bound())

    def lower_bound(self):
        return ndarray([.01, .01, .49, .01, .49, .01, .01, .01])

    def upper_bound(self):
        return ndarray([.49, .49, .99, .49, .99, .49, .49, .49])