-
Notifications
You must be signed in to change notification settings - Fork 0
/
control_frequency.py
200 lines (166 loc) · 6.68 KB
/
control_frequency.py
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
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
import get_param
import matplotlib.pyplot as plt
import matplotlib
from Logger import Logger,t_step
from pde_cnn import get_Net
import torch
from torch.optim import Adam
import numpy as np
from derivatives import vector2HSV,toCuda,toCpu,params,normal2staggered,staggered2normal,rot_mac
from torch.optim import Adam
import cv2
import math
import numpy as np
import time
torch.manual_seed(0)
torch.set_num_threads(4)
np.random.seed(0)
mu = params.mu
rho = params.rho
dt = params.dt
w,h = params.width,params.height
plot = True
n_warmup_time_steps = 500
n_time_steps=200
target_freq = min(max(params.target_freq,2),8)
print(f"target frequency: {target_freq}")
# load fluid model:
logger = Logger(get_param.get_hyperparam(params),use_csv=False,use_tensorboard=False)
pde_cnn = toCuda(get_Net(params))
date_time,index = logger.load_state(pde_cnn,None,datetime=params.load_date_time,index=params.load_index)
pde_cnn.eval()
print(f"loaded {params.net}: {date_time}, index: {index}")
# setup opencv windows for in depth visualizations
cv2.namedWindow('legend',cv2.WINDOW_NORMAL)
vector = toCuda(torch.cat([torch.arange(-1,1,0.01).unsqueeze(0).unsqueeze(2).repeat(1,1,200),torch.arange(-1,1,0.01).unsqueeze(0).unsqueeze(1).repeat(1,200,1)]))
image = vector2HSV(vector)
image = cv2.cvtColor(image,cv2.COLOR_HSV2BGR)
cv2.imshow('legend',image)
cv2.namedWindow('p',cv2.WINDOW_NORMAL)
cv2.namedWindow('v',cv2.WINDOW_NORMAL)
def torch_fft(input, signal_ndim, normalized=False):
if signal_ndim < 1 or signal_ndim > 3:
print("Signal ndim out of range, was", signal_ndim, "but expected a value between 1 and 3, inclusive")
return
dims = (-1)
if signal_ndim == 2:
dims = (-2, -1)
if signal_ndim == 3:
dims = (-3, -2, -1)
norm = "backward"
if normalized:
norm = "ortho"
return torch.view_as_real(torch.fft.fftn(torch.view_as_complex(input), dim=dims, norm=norm))
#generate environment for fluid simulation
def get_problem(w,h,object_x=50,object_y=50,object_w=5,object_h=10):
v_cond = toCuda(torch.zeros(1,2,h,w))
v_cond[0,1,10:(h-10),0:5]=1
v_cond[0,1,10:(h-10),(w-5):w]=1
cond_mask = toCuda(torch.zeros(1,1,h,w))
cond_mask[0,:,0:3,:]=1
cond_mask[0,:,(h-3):h,:]=1
cond_mask[0,:,:,0:5]=1
cond_mask[0,:,:,(w-5):w]=1
cond_mask[0,:,(object_y-object_h):(object_y+object_h),(object_x-object_w):(object_x+object_w)] = 1
flow_mask = 1-cond_mask
a_old = toCuda(torch.zeros(1,1,h,w))
p_old = toCuda(torch.zeros(1,1,h,w))
return v_cond,cond_mask,flow_mask,a_old,p_old
# initialize flow_v (this is the variable, we want to optimize such that we reach the target frequency)
start_v = 0.3
if params.cuda:
flow_v = torch.ones(1,1,1,1,requires_grad=True,device="cuda")
else:
flow_v = torch.ones(1,1,1,1,requires_grad=True)
# initialize optimizer
optim = Adam([flow_v],lr=0.2)
E_fft_ys = []
# to obtain smoother gradients, we scale the v_y(t) curve with a gaussian
velocity_y_curve_scaler = torch.exp(-((toCuda(torch.arange(n_time_steps).unsqueeze(1))-n_time_steps/2)/n_time_steps*4)**2)
# optimization loop:
for epoch in range(200):
v_cond,cond_mask,flow_mask,a_old,p_old = get_problem(w,h)
v_cond = normal2staggered(v_cond)
cond_mask_mac = (normal2staggered(cond_mask.repeat(1,2,1,1))==1).float()
flow_mask_mac = (normal2staggered(flow_mask.repeat(1,2,1,1))>=0.5).float()
# warm up simulation with n_warmup_time_steps
with torch.no_grad():
for t in range(n_warmup_time_steps):
a_new,p_new = pde_cnn(a_old,p_old,flow_mask,start_v*flow_v*v_cond,cond_mask)
p_new = (p_new-torch.mean(p_new,dim=(1,2,3)).unsqueeze(1).unsqueeze(2).unsqueeze(3))
a_new = (a_new-torch.mean(a_new,dim=(1,2,3)).unsqueeze(1).unsqueeze(2).unsqueeze(3))
a_old, p_old = a_new, p_new
# simulation that should be used for gradient propagation
velocity_y_curve = toCuda(torch.zeros(n_time_steps,2))
for t in range(n_time_steps):
a_new,p_new = pde_cnn(a_old,p_old,flow_mask,start_v*flow_v*v_cond,cond_mask)
p_new = (p_new-torch.mean(p_new,dim=(1,2,3)).unsqueeze(1).unsqueeze(2).unsqueeze(3))
a_new = (a_new-torch.mean(a_new,dim=(1,2,3)).unsqueeze(1).unsqueeze(2).unsqueeze(3))
v_new = rot_mac(a_new)
v_new = cond_mask_mac*v_cond*flow_v*start_v+flow_mask_mac*v_new
velocity_y_curve[t,0] = torch.mean(v_new[0,0:1,40:(h-40),80:100])
# visualize simulation progress
if t%10==0:
with torch.no_grad():
p = flow_mask[0,0]*p_new[0,0].clone()
p = p-torch.min(p)
p = p/torch.max(p)
cv2.imshow('p',toCpu(p).numpy())
vector = staggered2normal(v_new)[0].clone()
image = vector2HSV(vector.detach())
image = cv2.cvtColor(image,cv2.COLOR_HSV2BGR)
image[40:(h-40),80]=255
image[40:(h-40),100]=255
image[40,80:100]=255
image[(h-40),80:101]=255
cv2.imshow('v',image)
key = cv2.waitKey(1)
a_old, p_old = a_new, p_new
# compute expectation value of vortex shedding frequency
fft_y_curve = torch.sum(torch_fft(velocity_y_curve*velocity_y_curve_scaler,signal_ndim=1,normalized=True)**2,dim=1)
norm_fft_y_curve = fft_y_curve[:15]/torch.sum(fft_y_curve[:15])
E_fft_y = torch.sum(norm_fft_y_curve*torch.arange(0,15).cuda())
# loss function to push vortex shedding frequency towards target frequency
loss = (E_fft_y-target_freq)**2
# propagate gradients throughout simulation and optimize
optim.zero_grad()
loss.backward()
optim.step()
print(f"iteration {epoch}: E_fft_y = {E_fft_y}")
E_fft_ys.append(E_fft_y.detach().cpu().numpy())
# plot optimization progress
font = {'family':'normal','weight':'bold','size':22}
matplotlib.rc('font',**font)
plt.figure(1)
plt.clf()
plt.plot((velocity_y_curve)[:,0].detach().cpu().numpy(),linestyle="dashed",color="r")
plt.plot((velocity_y_curve*velocity_y_curve_scaler)[:,0].detach().cpu().numpy())
plt.draw()
plt.title(f"speed: {round(start_v*flow_v.detach().cpu().numpy()[0][0][0][0],3)}")
plt.xlabel("time")
plt.ylabel("$v_y$")
plt.legend(["$v_y(t)$","gaussian $\cdot v_y(t)$"])
plt.subplots_adjust(left=0.23,bottom=0.17)
plt.figure(2)
plt.clf()
fft_y_curve = fft_y_curve / torch.max(fft_y_curve)
plt.plot(fft_y_curve.detach().cpu().numpy()[:15])
plt.vlines(E_fft_y.detach().cpu().numpy(),0,1,color='r')
plt.vlines(target_freq,0,1,color='g')
plt.draw()
plt.title(f"speed: {round(start_v*flow_v.detach().cpu().numpy()[0][0][0][0],3)}")
plt.xlabel("frequency")
plt.ylabel("$|V_y|^2$")
plt.legend(["$|V_y(f)|^2$","$E[|V_y(f)|^2]$","target"])
plt.subplots_adjust(left=0.23,bottom=0.17)
plt.figure(3)
plt.clf()
plt.plot(E_fft_ys,color="r")
plt.hlines(target_freq,xmin=0,xmax=len(E_fft_ys)-1,color="g")
plt.legend(["$E[|V_y(f)|^2]$","target"])
plt.ylim((0,10))
plt.xlabel("iteration")
plt.ylabel("frequency")
plt.subplots_adjust(left=0.23,bottom=0.17)
plt.pause(0.0001)
del v_cond,cond_mask,flow_mask,a_new,p_new,a_old,p_old,velocity_y_curve,v_new