def mem_static_x_vector(mem, refn, vdc, type='bpt', atol=1e-10, maxiter=100):
'''
Static deflection of membrane calculated via fixed-point iteration.
'''
def pes(v, x, g_eff):
return -e_0 / 2 * v**2 / (g_eff + x)**2
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
K = mem_k_matrix(mem, refn, type=type)
g_eff = mem.gap + mem.isolation / mem.permittivity
F = mem_f_vector(mem, refn, 1)
Kinv = linalg.inv(K)
nnodes = K.shape[0]
x0 = np.zeros(nnodes)
for i in range(maxiter):
x0_new = Kinv.dot(F * pes(vdc, x0, g_eff))
if np.max(np.abs(x0_new - x0)) < atol:
is_collapsed = False
return x0_new, is_collapsed
x0 = x0_new
is_collapsed = True
return x0, is_collapsed
def mem_static_x_vector(mem, refn, vdc, k, x0, atol=1e-10, maxiter=100):
'''
'''
def pes(v, x, g_eff):
return -e_0 / 2 * v**2 / (g_eff + x)**2
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
K = fem.mem_k_matrix(mem, refn)
g_eff = mem.gap + mem.isolation / mem.permittivity
F = fem.mem_f_vector(mem, refn, 1)
Kinv = np.linalg.inv(K)
nnodes = K.shape[0]
x = np.zeros(nnodes)
for i in range(maxiter):
p = F * pes(vdc, x, g_eff)
p[x < x0] += -k * (x[x < x0] - x0) * F[x < x0]
xnew = Kinv.dot(p)
if np.max(np.abs(xnew - x)) < atol:
is_collapsed = False
return xnew, is_collapsed
x = xnew
is_collapsed = True
return x, p, is_collapsed
def mem_f_vector(mem, refn, p):
'''
Pressure load vector based on equal distribution of pressure to element
nodes.
'''
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
nodes = amesh.vertices
triangles = amesh.triangles
triangle_areas = amesh.g / 2
ob = amesh.on_boundary
f = np.zeros(len(nodes))
for tt in range(len(triangles)):
tri = triangles[tt, :]
ap = triangle_areas[tt]
bfac = 1 * np.sum(~ob[tri])
f[tri] += 1 / bfac * p * ap
f[ob] = 0
return f
def mem_dlm_matrix(mem, refn):
'''
Mass matrix based on equal distribution of element mass to nodes
(diagonally-lumped).
'''
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
# get mesh information
nodes = amesh.vertices
triangles = amesh.triangles
triangle_areas = amesh.g / 2
mass = sum([x * y for x, y in zip(mem.density, mem.thickness)])
# construct M matrix by adding contribution from each element
M = np.zeros((len(nodes), len(nodes)))
for tt in range(len(triangles)):
tri = triangles[tt, :]
ap = triangle_areas[tt]
M[tri, tri] += 1 / 3 * mass * ap
ob = amesh.on_boundary
M[ob, :] = 0
M[:, ob] = 0
M[ob, ob] = 1
return M
def mem_z_matrix(mem, refn, k, *args, **kwargs):
'''
Impedance matrix in FullFormat for a membrane.
'''
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
return np.array(ZFullMatrix(amesh, k, *args, **kwargs).data)
def calc_es_correction(array, refn):
fcorr = []
for elem in array.elements:
for mem in elem.membranes:
if isinstance(mem, abstract.SquareCmutMembrane):
square = True
amesh = mesh.square(mem.length_x, mem.length_y, refn=refn)
elif isinstance(mem, abstract.CircularCmutMembrane):
square = False
amesh = mesh.circle(mem.radius, refn=refn)
K = fem.mem_k_matrix(amesh, mem.y_modulus, mem.thickness, mem.p_ratio)
Kinv = np.linalg.inv(K)
g = mem.gap
g_eff = mem.gap + mem.isolation / mem.permittivity
# F = fem.mem_f_vector(amesh, 1)
F = np.array(fem.f_from_abstract(array, refn).todense())
# u = Kinv.dot(-F)
# unorm = u / np.max(np.abs(u))
for i, pat in enumerate(mem.patches):
if square:
avg = fem.square_patch_avg_vector(amesh.vertices, amesh.triangles, amesh.on_boundary,
mem.length_x, mem.length_y, pat.position[0] - mem.position[0],
pat.position[1] - mem.position[1], pat.length_x, pat.length_y)
else:
avg = fem.circular_patch_avg_vector(amesh.vertices, amesh.triangles, amesh.on_boundary,
mem.radius, pat.position[0] - mem.position[0], pat.position[1] - mem.position[1],
pat.radius_min, pat.radius_max, pat.theta_min, pat.theta_max)
u = Kinv.dot(-F[:,i])
unorm = u / np.max(np.abs(u))
d = np.linspace(0, 1, 11)
fc = []
uavg = []
fpp = []
for di in d:
ubar = (unorm * g * di).dot(avg) / pat.area
uavg.append(ubar)
fc.append((-e_0 / 2 / (unorm * g * di + g_eff)**2).dot(avg) / pat.area)
fpp.append(-e_0 / 2 / (ubar + g_eff)**2)
fcorr.append((d, unorm, uavg, fc, fpp))
# fcorr.append(interp1d(uavg, fc, kind='cubic', bounds_error=False, fill_value=(fc[-1], fc[0])))
return fcorr
def mem_f_vector_arb_load(mem, refn, load_func):
'''
Pressure load vector based on an arbitrary load.
'''
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
nodes = amesh.vertices
triangles = amesh.triangles
triangle_areas = amesh.triangle_areas
ob = amesh.on_boundary
f = np.zeros(len(nodes))
for tt in range(len(triangles)):
tri = triangles[tt, :]
xi, yi = nodes[tri[0], :2]
xj, yj = nodes[tri[1], :2]
xk, yk = nodes[tri[2], :2]
def load_func_psi_eta(psi, eta):
x = (xj - xi) * psi + (xk - xi) * eta + xi
y = (yj - yi) * psi + (yk - yi) * eta + yi
return load_func(x, y)
integ, _ = dblquad(load_func_psi_eta,
0,
1,
0,
lambda x: 1 - x,
epsrel=1e-1,
epsabs=1e-1)
frac = integ / (1 / 2) # fraction of triangle covered by load
da = triangle_areas[tt]
bfac = 1 * np.sum(~ob[tri])
f[tri] += 1 / bfac * frac * da
ob = amesh.on_boundary
f[ob] = 0
return f
def mem_eig(mem, refn):
'''
Returns the eigenfrequency (in Hz) and eigenmodes of a membrane.
'''
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
ob = amesh.on_boundary
M = mem_m_matrix(mem, refn, mu=0.5)
K = mem_k_matrix(mem, refn)
w, v = linalg.eig(linalg.inv(M).dot(K)[np.ix_(~ob, ~ob)])
idx = np.argsort(np.sqrt(np.abs(w)))
eigf = np.sqrt(np.abs(w))[idx] / (2 * np.pi)
eigv = v[:, idx]
return eigf, eigv
def mem_cm_matrix(mem, refn):
'''
Mass matrix based on kinetic energy and linear shape functions
(consistent).
'''
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
# get mesh information
nodes = amesh.vertices
triangles = amesh.triangles
triangle_areas = amesh.triangle_areas
mass = sum([x * y for x, y in zip(mem.density, mem.thickness)])
# construct M matrix by adding contribution from each element
M = np.zeros((len(nodes), len(nodes)))
for tt in range(len(triangles)):
tri = triangles[tt, :]
xi, yi = nodes[tri[0], :2]
xj, yj = nodes[tri[1], :2]
xk, yk = nodes[tri[2], :2]
# da = ((xj - x5) * (yk - yi) - (xk - x5) * (yj - yi))
da = triangle_areas[tt]
Mt = np.array([[1, 1 / 2, 1 / 2], [1 / 2, 1, 1 / 2], [1 / 2, 1 / 2, 1]
]) / 12
M[np.ix_(tri, tri)] += 2 * Mt * mass * da
ob = amesh.on_boundary
M[ob, :] = 0
M[:, ob] = 0
M[ob, ob] = 1
return M
def mem_k_matrix_bpt(mem, refn):
'''
Stiffness matrix based on 3-dof (rotation-free) basic plate triangle (BPT) elements.
Refer to E. Onate and F. Zarate, Int. J. Numer. Meth. Engng. 47, 557-603 (2000).
'''
def L(x1, y1, x2, y2):
# calculates edge length
return np.sqrt((x2 - x1)**2 + (y2 - y1)**2)
def T_matrix(x1, y1, x2, y2):
# transformation matrix for an edge
z = [0, 0, 1]
r = [x2 - x1, y2 - y1]
n = np.cross(z, r)
norm = np.linalg.norm(n)
n = n / norm
nx, ny, _ = n
return np.array([[-nx, 0], [0, -ny], [-ny, -nx]])
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
# get mesh information
nodes = amesh.vertices
triangles = amesh.triangles
triangle_edges = amesh.triangle_edges
triangle_areas = amesh.triangle_areas
ntriangles = len(triangles)
ob = amesh.on_boundary
# determine list of neighbors for each triangle
# None indicates neighbor doesn't exist for that edge (boundary edge)
triangle_neighbors = []
for tt in range(ntriangles):
neighbors = []
for te in triangle_edges[tt, :]:
mask = np.any(triangle_edges == te, axis=1)
args = np.nonzero(mask)[0]
if len(args) > 1:
neighbors.append(args[args != tt][0])
else:
neighbors.append(None)
triangle_neighbors.append(neighbors)
amesh.triangle_neighbors = triangle_neighbors
# construct constitutive matrix for material
h = mem.thickness[0] # no support for composite membranes yet
E = mem.y_modulus[0]
eta = mem.p_ratio[0]
D = np.zeros((3, 3))
D[0, 0] = 1
D[0, 1] = eta
D[1, 0] = eta
D[1, 1] = 1
D[2, 2] = (1 - eta) / 2
D = D * E * h**3 / (12 * (1 - eta**2))
# calculate Jacobian and gradient operator for each triangle
gradops = []
for tt in range(ntriangles):
tri = triangles[tt, :]
xi, yi = nodes[tri[0], :2]
xj, yj = nodes[tri[1], :2]
xk, yk = nodes[tri[2], :2]
J = np.array([[xj - xi, xk - xi], [yj - yi, yk - yi]])
gradop = np.linalg.inv(J.T).dot([[-1, 1, 0], [-1, 0, 1]])
# gradop = np.linalg.inv(J.T).dot([[1, 0, -1],[0, 1, -1]])
gradops.append(gradop)
# construct K matrix
K = np.zeros((len(nodes), len(nodes)))
for p in range(ntriangles):
trip = triangles[p, :]
ap = triangle_areas[p]
xi, yi = nodes[trip[0], :2]
xj, yj = nodes[trip[1], :2]
xk, yk = nodes[trip[2], :2]
neighbors = triangle_neighbors[p]
gradp = gradops[p]
# list triangle edges, ordered so that z cross-product will produce outward normal
edges = [(xk, yk, xj, yj), (xi, yi, xk, yk), (xj, yj, xi, yi)]
# begin putting together indexes needed later for matrix assignment
ii, jj, kk = trip
Kidx = [ii, jj, kk]
Kpidx = [0, 1, 2]
# construct B matrix for control element
Bp = np.zeros((3, 6))
for j, n in enumerate(neighbors):
if n is None:
continue
# determine index of the node in the neighbor opposite edge
iin, jjn, kkn = triangles[n, :]
uidx = [
x for x in np.unique([ii, jj, kk, iin, jjn, kkn])
if x not in [ii, jj, kk]
][0]
# update indexes
Kidx.append(uidx)
Kpidx.append(3 + j)
l = L(*edges[j])
T = T_matrix(*edges[j])
gradn = gradops[n]
pterm = l / 2 * T.dot(gradp)
Bp[:, :3] += pterm
nterm = l / 2 * T.dot(gradn)
idx = [Kpidx[Kidx.index(x)] for x in [iin, jjn, kkn]]
Bp[:, idx] += nterm
Bp = Bp / ap
# construct local K matrix for control element
Kp = (Bp.T).dot(D).dot(Bp) * ap
# add matrix values to global K matrix
K[np.ix_(Kidx, Kidx)] += Kp[np.ix_(Kpidx, Kpidx)]
K[ob, :] = 0
K[:, ob] = 0
K[ob, ob] = 1
return K
def mem_k_matrix_hybrid(mem, refn):
'''
*Experimental* Stiffness matrix based on HPB for interior triangles and
BPT for boundary triangles.
'''
def norm(r1, r2):
# calculates edge length
return np.sqrt((r2[0] - r1[0])**2 + (r2[1] - r1[1])**2)
def LL(x1, y1, x2, y2):
# calculates edge length
return np.sqrt((x2 - x1)**2 + (y2 - y1)**2)
def T_matrix(x1, y1, x2, y2):
# transformation matrix for an edge
z = [0, 0, 1]
r = [x2 - x1, y2 - y1]
n = np.cross(z, r)
norm = np.linalg.norm(n)
n = n / norm
nx, ny, _ = n
return np.array([[-nx, 0], [0, -ny], [-ny, -nx]])
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
# get mesh information
nodes = amesh.vertices
triangles = amesh.triangles
triangle_edges = amesh.triangle_edges
triangle_areas = amesh.triangle_areas
ntriangles = len(triangles)
ob = amesh.on_boundary
# determine list of neighbors and neighbor nodes for each triangle
# None indicates neighbor doesn't exist for that edge (boundary edge)
triangle_neighbors = []
neighbors_node = []
for tt in range(ntriangles):
neighbors = []
neighbor_node = []
for te in triangle_edges[tt, :]:
mask = np.any(triangle_edges == te, axis=1)
args = np.nonzero(mask)[0]
if len(args) > 1:
n = args[args != tt][0]
neighbors.append(n)
# determine index of the node in the neighbor opposite edge
iin, jjn, kkn = triangles[n, :]
ii, jj, kk = triangles[tt, :]
uidx = [
x for x in np.unique([ii, jj, kk, iin, jjn, kkn])
if x not in [ii, jj, kk]
][0]
neighbor_node.append(uidx)
else:
neighbors.append(None)
neighbor_node.append(None)
triangle_neighbors.append(neighbors)
neighbors_node.append(neighbor_node)
amesh.triangle_neighbors = triangle_neighbors
amesh.neighbors_node = neighbors_node
# construct constitutive matrix for material
h = mem.thickness[0] # no support for composite membranes yet
E = mem.y_modulus[0]
eta = mem.p_ratio[0]
D = np.zeros((3, 3))
D[0, 0] = 1
D[0, 1] = eta
D[1, 0] = eta
D[1, 1] = 1
D[2, 2] = (1 - eta)
D = D * E * h**3 / (12 * (1 - eta**2))
D2 = np.zeros((3, 3))
D2[0, 0] = 1
D2[0, 1] = eta
D2[1, 0] = eta
D2[1, 1] = 1
D2[2, 2] = (1 - eta) / 2
D2 = D2 * E * h**3 / (12 * (1 - eta**2))
# calculate Jacobian and gradient operator for each triangle
gradops = []
for tt in range(ntriangles):
tri = triangles[tt, :]
xi, yi = nodes[tri[0], :2]
xj, yj = nodes[tri[1], :2]
xk, yk = nodes[tri[2], :2]
J = np.array([[xj - xi, xk - xi], [yj - yi, yk - yi]])
gradop = np.linalg.inv(J.T).dot([[-1, 1, 0], [-1, 0, 1]])
gradops.append(gradop)
# construct K matrix
K = np.zeros((len(nodes), len(nodes)))
for p in range(ntriangles):
trip = triangles[p, :]
ap = triangle_areas[p]
# assign primary triangle nodes based on order of edges
x5, y5 = nodes[trip[0], :2]
x6, y6 = nodes[trip[1], :2]
x4, y4 = nodes[trip[2], :2]
r4 = np.array([x4, y4])
r5 = np.array([x5, y5])
r6 = np.array([x6, y6])
# assign neighboring triangle nodes and add fictitious nodes if necessary
neighbors = neighbors_node[p]
if None in neighbors or ob[trip[0]] or ob[trip[1]] or ob[trip[2]]:
# if True:
xi, yi = nodes[trip[0], :2]
xj, yj = nodes[trip[1], :2]
xk, yk = nodes[trip[2], :2]
neighbors = triangle_neighbors[p]
gradp = gradops[p]
# list triangle edges, ordered so that z cross-product will produce outward normal
edges = [(xk, yk, xj, yj), (xi, yi, xk, yk), (xj, yj, xi, yi)]
# begin putting together indexes needed later for matrix assignment
ii, jj, kk = trip
Kidx = [ii, jj, kk]
Kpidx = [0, 1, 2]
# construct B matrix for control element
Bp = np.zeros((3, 6))
for j, n in enumerate(neighbors):
if n is None:
continue
# determine index of the node in the neighbor opposite edge
iin, jjn, kkn = triangles[n, :]
uidx = [
x for x in np.unique([ii, jj, kk, iin, jjn, kkn])
if x not in [ii, jj, kk]
][0]
# update indexes
Kidx.append(uidx)
Kpidx.append(3 + j)
l = LL(*edges[j])
T = T_matrix(*edges[j])
gradn = gradops[n]
pterm = l / 2 * T.dot(gradp)
Bp[:, :3] += pterm
nterm = l / 2 * T.dot(gradn)
idx = [Kpidx[Kidx.index(x)] for x in [iin, jjn, kkn]]
Bp[:, idx] += nterm
Bp = Bp / ap
# construct local K matrix for control element
Kp = (Bp.T).dot(D).dot(Bp) * ap
# add matrix values to global K matrix
K[np.ix_(Kidx, Kidx)] += Kp[np.ix_(Kpidx, Kpidx)]
else:
x1, y1 = nodes[neighbors[0], :2]
x2, y2 = nodes[neighbors[1], :2]
x3, y3 = nodes[neighbors[2], :2]
r1 = np.array([x1, y1])
r2 = np.array([x2, y2])
r3 = np.array([x3, y3])
# construct C matrix
C = np.zeros((6, 3))
C[0, :] = [x1**2 / 2, y1**2 / 2, x1 * y1]
C[1, :] = [x2**2 / 2, y2**2 / 2, x2 * y2]
C[2, :] = [x3**2 / 2, y3**2 / 2, x3 * y3]
C[3, :] = [x4**2 / 2, y4**2 / 2, x4 * y4]
C[4, :] = [x5**2 / 2, y5**2 / 2, x5 * y5]
C[5, :] = [x6**2 / 2, y6**2 / 2, x6 * y6]
# construct L matrix
# calculate vars based on geometry of first sub-element
L1 = norm(r6, r4)
b1a1 = (r1 - r4).dot(r6 - r4) / L1
b1a2 = (r1 - r6).dot(r4 - r6) / L1
b1b1 = (r5 - r4).dot(r6 - r4) / L1
b1b2 = (r5 - r6).dot(r4 - r6) / L1
h1a = np.sqrt(norm(r1, r4)**2 - b1a1**2)
h1b = np.sqrt(norm(r5, r6)**2 - b1b2**2)
# calculate vars based on geometry of second sub-element
L2 = norm(r5, r4)
b2a1 = (r2 - r5).dot(r4 - r5) / L2
b2a2 = (r2 - r4).dot(r5 - r4) / L2
b2b1 = (r6 - r5).dot(r4 - r5) / L2
b2b2 = (r6 - r4).dot(r5 - r4) / L2
h2a = np.sqrt(norm(r2, r5)**2 - b2a1**2)
h2b = np.sqrt(norm(r6, r4)**2 - b2b2**2)
# calculate vars based on geometry of third sub-element
L3 = norm(r5, r6)
b3a1 = (r3 - r6).dot(r5 - r6) / L3
b3a2 = (r3 - r5).dot(r6 - r5) / L3
b3b1 = (r4 - r6).dot(r5 - r6) / L3
b3b2 = (r4 - r5).dot(r6 - r5) / L3
h3a = np.sqrt(norm(r3, r6)**2 - b3a1**2)
h3b = np.sqrt(norm(r4, r5)**2 - b3b2**2)
L = np.zeros((3, 6))
L[0, 0] = 1 / h1a
L[1, 1] = 1 / h2a
L[2, 2] = 1 / h3a
L[0, 3] = -b1a2 / (L1 * h1a) + -b1b2 / (L1 * h1b)
L[0, 4] = 1 / h1b
L[0, 5] = -b1a1 / (L1 * h1a) + -b1b1 / (L1 * h1b)
L[1, 3] = -b2b1 / (L2 * h2b) + -b2a1 / (L2 * h2a)
L[1, 4] = -b2b2 / (L2 * h2b) + -b2a2 / (L2 * h2a)
L[1, 5] = 1 / h2b
L[2, 3] = 1 / h3b
L[2, 4] = -b3a1 / (L3 * h3a) + -b3b1 / (L3 * h3b)
L[2, 5] = -b3b2 / (L3 * h3b) + -b3a2 / (L3 * h3a)
# calculate G matrix
G = L @ C
Ginv = np.linalg.inv(G)
# create I_D matrix
I_D = np.zeros((3, 3))
I_D[0, 0] = 1
I_D[1, 1] = 1
I_D[2, 2] = 2
# K_be = ap * (L.T).dot(Ginv.T).dot(I_D).dot(D).dot(Ginv).dot(L)
K_be = ap * L.T @ Ginv.T @ I_D @ D @ Ginv @ L
# begin putting together indexes needed later for matrix assignment
K_idx = [
neighbors[0], neighbors[1], neighbors[2], trip[2], trip[0],
trip[1]
]
K_be_idx = [0, 1, 2, 3, 4, 5]
# add matrix values to global K matrix
K[np.ix_(K_idx, K_idx)] += K_be[np.ix_(K_be_idx, K_be_idx)]
K[ob, :] = 0
K[:, ob] = 0
K[ob, ob] = 1
return K
def mem_k_matrix_hpb(mem, refn, retmesh=False):
'''
Stiffness matrix based on 3-dof hinged plate bending (HPB) elements.
Refer to R. Phaal and C. R. Calladine, Int. J. Numer. Meth. Engng.,
vol. 35, no. 5, pp. 955–977, (1992).
'''
def norm(r1, r2):
# calculates edge length
return np.sqrt((r2[0] - r1[0])**2 + (r2[1] - r1[1])**2)
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x, mem.length_y, refn)
else:
amesh = mesh.circle(mem.radius, refn)
# get mesh information
nodes = amesh.vertices
triangles = amesh.triangles
triangle_edges = amesh.triangle_edges
triangle_areas = amesh.triangle_areas
ntriangles = len(triangles)
ob = amesh.on_boundary
# determine list of neighbors and neighbor nodes for each triangle
# None indicates neighbor doesn't exist for that edge (boundary edge)
triangle_neighbors = []
neighbors_node = []
for tt in range(ntriangles):
neighbors = []
neighbor_node = []
for te in triangle_edges[tt, :]:
mask = np.any(triangle_edges == te, axis=1)
args = np.nonzero(mask)[0]
if len(args) > 1:
n = args[args != tt][0]
neighbors.append(n)
# determine index of the node in the neighbor opposite edge
iin, jjn, kkn = triangles[n, :]
ii, jj, kk = triangles[tt, :]
uidx = [
x for x in np.unique([ii, jj, kk, iin, jjn, kkn])
if x not in [ii, jj, kk]
][0]
neighbor_node.append(uidx)
else:
neighbors.append(None)
neighbor_node.append(None)
triangle_neighbors.append(neighbors)
neighbors_node.append(neighbor_node)
amesh.triangle_neighbors = triangle_neighbors
amesh.neighbors_node = neighbors_node
# construct constitutive matrix for material
h = mem.thickness[0] # no support for composite membranes yet
E = mem.y_modulus[0]
eta = mem.p_ratio[0]
D = np.zeros((3, 3))
D[0, 0] = 1
D[0, 1] = eta
D[1, 0] = eta
D[1, 1] = 1
D[2, 2] = (1 - eta)
D = D * E * h**3 / (12 * (1 - eta**2))
# construct K matrix
K = np.zeros((len(nodes), len(nodes)))
for p in range(ntriangles):
trip = triangles[p, :]
ap = triangle_areas[p]
# assign primary triangle nodes based on order of edges
x5, y5 = nodes[trip[0], :2]
x6, y6 = nodes[trip[1], :2]
x4, y4 = nodes[trip[2], :2]
r4 = np.array([x4, y4])
r5 = np.array([x5, y5])
r6 = np.array([x6, y6])
# assign neighboring triangle nodes and add fictitious nodes if necessary
neighbors = neighbors_node[p]
if neighbors[0] is None:
x = r5
xo = r6
n = (r4 - r6) / norm(r4, r6)
x1, y1 = -x + 2 * xo + 2 * (x - xo).dot(n) * n
else:
x1, y1 = nodes[neighbors[0], :2]
if neighbors[1] is None:
x = r6
xo = r4
n = (r5 - r4) / norm(r5, r4)
x2, y2 = -x + 2 * xo + 2 * (x - xo).dot(n) * n
else:
x2, y2 = nodes[neighbors[1], :2]
if neighbors[2] is None:
x = r4
xo = r5
n = (r6 - r5) / norm(r6, r5)
x3, y3 = -x + 2 * xo + 2 * (x - xo).dot(n) * n
else:
x3, y3 = nodes[neighbors[2], :2]
r1 = np.array([x1, y1])
r2 = np.array([x2, y2])
r3 = np.array([x3, y3])
# construct C matrix
C = np.zeros((6, 3))
C[0, :] = [x1**2 / 2, y1**2 / 2, x1 * y1]
C[1, :] = [x2**2 / 2, y2**2 / 2, x2 * y2]
C[2, :] = [x3**2 / 2, y3**2 / 2, x3 * y3]
C[3, :] = [x4**2 / 2, y4**2 / 2, x4 * y4]
C[4, :] = [x5**2 / 2, y5**2 / 2, x5 * y5]
C[5, :] = [x6**2 / 2, y6**2 / 2, x6 * y6]
# construct L matrix
# calculate vars based on geometry of first sub-element
L1 = norm(r6, r4)
b1a1 = (r1 - r4).dot(r6 - r4) / L1
b1a2 = (r1 - r6).dot(r4 - r6) / L1
b1b1 = (r5 - r4).dot(r6 - r4) / L1
b1b2 = (r5 - r6).dot(r4 - r6) / L1
h1a = np.sqrt(norm(r1, r4)**2 - b1a1**2)
h1b = np.sqrt(norm(r5, r6)**2 - b1b2**2)
# calculate vars based on geometry of second sub-element
L2 = norm(r5, r4)
b2a1 = (r2 - r5).dot(r4 - r5) / L2
b2a2 = (r2 - r4).dot(r5 - r4) / L2
b2b1 = (r6 - r5).dot(r4 - r5) / L2
b2b2 = (r6 - r4).dot(r5 - r4) / L2
h2a = np.sqrt(norm(r2, r5)**2 - b2a1**2)
h2b = np.sqrt(norm(r6, r4)**2 - b2b2**2)
# calculate vars based on geometry of third sub-element
L3 = norm(r5, r6)
b3a1 = (r3 - r6).dot(r5 - r6) / L3
b3a2 = (r3 - r5).dot(r6 - r5) / L3
b3b1 = (r4 - r6).dot(r5 - r6) / L3
b3b2 = (r4 - r5).dot(r6 - r5) / L3
h3a = np.sqrt(norm(r3, r6)**2 - b3a1**2)
h3b = np.sqrt(norm(r4, r5)**2 - b3b2**2)
L = np.zeros((3, 6))
L[0, 0] = 1 / h1a
L[1, 1] = 1 / h2a
L[2, 2] = 1 / h3a
L[0, 3] = -b1a2 / (L1 * h1a) + -b1b2 / (L1 * h1b)
L[0, 4] = 1 / h1b
L[0, 5] = -b1a1 / (L1 * h1a) + -b1b1 / (L1 * h1b)
L[1, 3] = -b2b1 / (L2 * h2b) + -b2a1 / (L2 * h2a)
L[1, 4] = -b2b2 / (L2 * h2b) + -b2a2 / (L2 * h2a)
L[1, 5] = 1 / h2b
L[2, 3] = 1 / h3b
L[2, 4] = -b3a1 / (L3 * h3a) + -b3b1 / (L3 * h3b)
L[2, 5] = -b3b2 / (L3 * h3b) + -b3a2 / (L3 * h3a)
# calculate G matrix
G = L @ C
Ginv = np.linalg.inv(G)
# create I_D matrix
I_D = np.zeros((3, 3))
I_D[0, 0] = 1
I_D[1, 1] = 1
I_D[2, 2] = 2
# K_be = ap * (L.T).dot(Ginv.T).dot(I_D).dot(D).dot(Ginv).dot(L)
K_be = ap * L.T @ Ginv.T @ I_D @ D @ Ginv @ L
# begin putting together indexes needed later for matrix assignment
K_idx = [trip[2], trip[0], trip[1]]
K_be_idx = [3, 4, 5]
# apply BCs
if neighbors[0] is None:
# fictitious node index = 0
# mirrored node index = 4
# boundary nodes index = 3, 5
# non-boundary nodes index = 1, 2
# modify row for mirrored node
K_be[4,
4] = (K_be[0, 0] + K_be[0, 4] + K_be[4, 0] + K_be[4, 4]) #/ 2
K_be[4, 1] = (K_be[4, 1] + K_be[0, 1]) #/ 2
K_be[4, 2] = (K_be[4, 2] + K_be[0, 2]) #/ 2
# K_be[4, 1] /= 2
# K_be[1, 4] /= 2
# K_be[4, 2] /= 2
# K_be[2, 4] /= 2
# K_be[4, 3] = (K_be[4, 3] + K_be[0, 3]) / 2
# K_be[4, 5] = (K_be[4, 5] + K_be[0, 5]) / 2
# modify row of first non-boundary node
K_be[1, 4] = (K_be[1, 4] + K_be[1, 0]) #/ 2
# modify row of second non-boundary node
K_be[2, 4] = (K_be[2, 4] + K_be[2, 0]) #/ 2
# modify row of first boundary node
# K_be[3, 3] = K_be[3, 3] / 2
# K_be[3, 4] = (K_be[3, 4] + K_be[3, 0]) / 2
# modify row of second boundary node
# K_be[5, 5] = K_be[5, 5] / 2
# K_be[5, 4] = (K_be[5, 4] + K_be[5, 0]) / 2
else:
K_idx.append(neighbors[0])
K_be_idx.append(0)
if neighbors[1] is None:
# fictitious node index = 1
# mirrored node index = 5
# boundary nodes index = 3, 4
# non-boundary nodes index = 0, 2
# modify row for mirrored node
K_be[5,
5] = (K_be[1, 1] + K_be[1, 5] + K_be[5, 1] + K_be[5, 5]) #/ 2
K_be[5, 0] = (K_be[5, 0] + K_be[1, 0]) #/ 2
K_be[5, 2] = (K_be[5, 2] + K_be[1, 2]) #/ 2
# K_be[5, 0] /= 2
# K_be[0, 5] /= 2
# K_be[5, 2] /= 2
# K_be[2, 5] /= 2
# K_be[5, 3] = (K_be[5, 3] + K_be[1, 3]) / 2
# K_be[5, 4] = (K_be[5, 4] + K_be[1, 4]) / 2
# modify row of first non-boundary node
K_be[0, 5] = (K_be[0, 5] + K_be[0, 1]) #/ 2
# modify row of second non-boundary node
K_be[2, 5] = (K_be[2, 5] + K_be[2, 1]) #/ 2
# modify row of first boundary node
# K_be[3, 3] = K_be[3, 3] / 2
# K_be[3, 5] = (K_be[3, 5] + K_be[3, 1]) / 2
# modify row of second boundary node
# K_be[4, 4] = K_be[4, 4] / 2
# K_be[4, 5] = (K_be[4, 5] + K_be[4, 1]) / 2
else:
K_idx.append(neighbors[1])
K_be_idx.append(1)
if neighbors[2] is None:
# fictitious node index = 2
# mirrored node index = 3
# boundary nodes index = 4, 5
# non-boundary nodes index = 0, 1
# modify row for mirrored node
K_be[3,
3] = (K_be[2, 2] + K_be[2, 3] + K_be[3, 2] + K_be[3, 3]) #/ 2
K_be[3, 0] = (K_be[3, 0] + K_be[2, 0]) #/ 2
K_be[3, 1] = (K_be[3, 1] + K_be[2, 1]) #/ 2
# K_be[3, 0] /= 2
# K_be[0, 3] /= 2
# K_be[3, 1] /= 2
# K_be[1, 3] /= 2
# K_be[3, 4] = (K_be[3, 4] + K_be[2, 4]) / 2
# K_be[3, 5] = (K_be[3, 5] + K_be[2, 5]) / 2
# modify row of first non-boundary node
K_be[0, 3] = (K_be[0, 3] + K_be[0, 2]) #/ 2
# modify row of second non-boundary node
K_be[1, 3] = (K_be[1, 3] + K_be[1, 2]) #/ 2
# modify row of first boundary node
# K_be[4, 4] = K_be[4, 4] / 2
# K_be[4, 3] = (K_be[4, 3] + K_be[4, 2]) / 2
# modify row of second boundary node
# K_be[5, 5] = K_be[5, 5] / 2
# K_be[5, 3] = (K_be[5, 3] + K_be[5, 2]) / 2
else:
K_idx.append(neighbors[2])
K_be_idx.append(2)
# add matrix values to global K matrix
K[np.ix_(K_idx, K_idx)] += K_be[np.ix_(K_be_idx, K_be_idx)]
K[ob, :] = 0
K[:, ob] = 0
K[ob, ob] = 1
return K
def mem_patch_avg_matrix(mem, refn):
'''
Averaging vector for a patch.
'''
if isinstance(mem, abstract.SquareCmutMembrane):
amesh = mesh.square(mem.length_x,
mem.length_y,
refn,
center=mem.position)
else:
amesh = mesh.circle(mem.radius, refn, center=mem.position)
nodes = amesh.vertices
triangles = amesh.triangles
triangle_areas = amesh.triangle_areas
# ob = amesh.on_boundary
avg = []
for pat in mem.patches:
if isinstance(mem, abstract.SquareCmutMembrane):
px, py, pz = pat.position
plx = pat.length_x
ply = pat.length_y
def load_func(x, y):
# use 2 * eps to account for potential round-off error
if x - (px - plx / 2) >= -2 * eps:
if x - (px + plx / 2) <= 2 * eps:
if y - (py - ply / 2) >= -2 * eps:
if y - (py + ply / 2) <= 2 * eps:
return 1
return 0
else:
px, py, pz = pat.position
prmin = pat.radius_min
prmax = pat.radius_max
pthmin = pat.theta_min
pthmax = pat.theta_max
def load_func(x, y):
r = np.sqrt((x - px)**2 + (y - py)**2)
th = np.arctan2((y - py), (x - px))
# pertube theta by 2 * eps to account for potential round-off error
th1 = th - 2 * eps
if th1 < -np.pi:
th1 += 2 * np.pi # account for [-pi, pi] wrap-around
th2 = th + 2 * eps
if th2 > np.pi:
th2 -= 2 * np.pi # account for [-pi, pi] wrap-around
if r - prmin >= -2 * eps:
if r - prmax <= 2 * eps:
# for theta, check both perturbed values
if th1 - pthmin >= 0:
if th1 - pthmax <= 0:
return 1
if th2 - pthmin >= 0:
if th2 - pthmax <= 0:
return 1
return 0
avg_pat = np.zeros(len(nodes))
for tt in range(len(triangles)):
tri = triangles[tt, :]
xi, yi = nodes[tri[0], :2]
xj, yj = nodes[tri[1], :2]
xk, yk = nodes[tri[2], :2]
# check if triangle vertices are inside or outside load
loadi = load_func(xi, yi)
loadj = load_func(xj, yj)
loadk = load_func(xk, yk)
# if load covers entire triangle
if all([loadi, loadj, loadk]):
da = triangle_areas[tt]
avg_pat[tri] += 1 / 3 * 1 * da
# if load does not cover any part of triangle
elif not any([loadi, loadj, loadk]):
continue
# if load partially covers triangle
else:
def load_func_psi_eta(psi, eta):
x = (xj - xi) * psi + (xk - xi) * eta + xi
y = (yj - yi) * psi + (yk - yi) * eta + yi
return load_func(x, y)
integ, _ = dblquad(load_func_psi_eta,
0,
1,
0,
lambda x: 1 - x,
epsrel=1e-1,
epsabs=1e-1)
frac = integ / (1 / 2) # fraction of triangle covered by load
da = triangle_areas[tt]
avg_pat[tri] += 1 / 3 * frac * da
avg.append(avg_pat / pat.area)
return np.array(avg).T