def m_model(input_location, input_scale):
dLInputScaling = getScalingDenseLayer(input_location, input_scale)
L1 = Dense(10)
L2 = Dense(5)
L3 = Dense(1, activation='linear', trainable=True)
model = Sequential([dLInputScaling, L1, L2, L3], name='m_mlp')
optimizer = RMSprop(1e-3)
model.compile(loss='mse', optimizer=optimizer, metrics=['mae'])
return model
def logC_model(input_location, input_scale):
dLInputScaling = getScalingDenseLayer(input_location, input_scale)
L1 = Dense(40, activation='elu')
L2 = Dense(20, activation='elu')
L3 = Dense(10, activation='elu')
L4 = Dense(5, activation='elu')
L5 = Dense(1, activation='elu', trainable=True)
model = Sequential([dLInputScaling, L1, L2, L3, L4, L5], name='c_mlp')
optimizer = RMSprop(1e-3)
model.compile(loss='mse', optimizer=optimizer, metrics=['mae'])
return model
def arch_model(switch, input_location, input_scale):
dLInputScaling = getScalingDenseLayer(input_location, input_scale)
if switch == 1:
L1 = Dense(5, activation = 'tanh')
L2 = Dense(1, activation = 'linear')
model = Sequential([dLInputScaling,L1,L2], name = 'bias_mlp')
model.compile(loss='mse', optimizer=RMSprop(1e-3), metrics=['mae'])
elif switch == 2:
L1 = Dense(10, activation = 'elu')
L2 = Dense(5, activation = 'elu')
L3 = Dense(1, activation = 'linear')
model = Sequential([dLInputScaling,L1,L2,L3], name = 'bias_mlp')
model.compile(loss='mse', optimizer=RMSprop(1e-3), metrics=['mae'])
elif switch == 3:
L1 = Dense(10, activation = 'elu')
L2 = Dense(5, activation = 'sigmoid')
L3 = Dense(1, activation = 'elu')
model = Sequential([dLInputScaling,L1,L2,L3], name = 'bias_mlp')
model.compile(loss='mse', optimizer=RMSprop(1e-3), metrics=['mae'])
elif switch == 4:
L1 = Dense(10, activation = 'tanh')
L2 = Dense(5, activation = 'tanh')
L3 = Dense(1, activation = 'elu')
model = Sequential([dLInputScaling,L1,L2,L3], name = 'bias_mlp')
model.compile(loss='mse', optimizer=RMSprop(1e-3), metrics=['mae'])
elif switch == 5:
L1 = Dense(20, activation = 'tanh')
L2 = Dense(10, activation = 'elu')
L3 = Dense(5, activation = 'sigmoid')
L4 = Dense(1, activation = 'linear', trainable = True)
model = Sequential([dLInputScaling,L1,L2,L3,L4], name = 'bias_mlp')
model.compile(loss='mse', optimizer=RMSprop(1e-3), metrics=['mae'])
elif switch == 6:
L1 = Dense(20, activation = 'elu')
L2 = Dense(10, activation = 'sigmoid')
L3 = Dense(5, activation = 'sigmoid')
L4 = Dense(1, activation = 'elu', trainable = True)
model = Sequential([dLInputScaling,L1,L2,L3,L4], name = 'bias_mlp')
model.compile(loss='mse', optimizer=RMSprop(1e-3), metrics=['mae'])
elif switch == 7:
L1 = Dense(40, activation = 'elu')
L2 = Dense(20, activation = 'sigmoid')
L3 = Dense(10, activation = 'sigmoid')
L4 = Dense(1, activation = 'elu', trainable = True)
model = Sequential([dLInputScaling,L1,L2,L3,L4], name = 'bias_mlp')
model.compile(loss='mse', optimizer=RMSprop(1e-3), metrics=['mae'])
return model
Dense(1,activation = 'sigmoid')
], name='plane_delgrs_mlp')
optimizer = RMSprop(0.01)
model.compile(loss='mean_squared_error',
optimizer=optimizer,
metrics=['mean_absolute_error', 'mean_squared_error'])
return model
parent_dir = os.path.dirname(os.getcwd())
dfPlane = pd.read_csv(parent_dir+'\data\\random_plane_set_500_adv.csv')
inputsMLPTrain = dfPlane[['Dkappa','dynamicLoads','bearingTemp']]
inputsMLPTrain_min = inputsMLPTrain.min(axis=0)
inputsMLPTrain_range = inputsMLPTrain.max(axis=0) - inputsMLPTrain_min
dLInputScaling = getScalingDenseLayer(inputsMLPTrain_min,inputsMLPTrain_range)
MLPmodel = build_model(dLInputScaling)
outputsMLPTrain = dfPlane[['delDkappa']]
outputsMLPTrain_min = outputsMLPTrain.min(axis=0)
outputsMLPTrain_range = outputsMLPTrain.max(axis=0) - outputsMLPTrain_min
outputsMLPTrain_norm = (outputsMLPTrain - outputsMLPTrain_min)/outputsMLPTrain_range
dfTrueDOE = pd.read_csv(parent_dir+'\data\\true_set_500_adv.csv')
inputsMLPPred = dfTrueDOE[['Dkappa','dynamicLoads','bearingTemp']]
outputsMLPPred = dfTrueDOE[['delDkappa']]
outputsMLPPred_min = outputsMLPPred.min(axis=0)
outputsMLPPred_range = outputsMLPPred.max(axis=0) - outputsMLPPred_min
plastic_io = pd.read_csv('./plastic_deflections2.csv',
index_col=False,
header=None)
force_input = np.asarray(plastic_io)[0, :]
plastic_deflections = np.asarray(plastic_io)[1:, :]
F_matrix[-1] += -NBC[-1]
force_vector = F_matrix
physics_model = create_physics_model(np.array([A_matrix]), force_vector, 2e5,
6e5, force_input.shape, 'float32')
elastic_deflection = physics_model.predict(force_input)
dLInputScaling = getScalingDenseLayer(
np.array([force_input.min(axis=0)]),
np.array([force_input.max(axis=0) - force_input.min(axis=0)]))
delta_stiffness_mlp = build_mlp(dLInputScaling, 'delta_stiffness')
delta_stiffness_mlp.trainable = True
force_mlp = build_force_mlp(dLInputScaling, 'delta_force')
force_mlp.trainable = True
fe_model = create_fe_model(force_mlp, delta_stiffness_mlp,
np.array([A_matrix]), force_vector, 1e1, 1e5, -1e3,
1e3, force_input.shape, 'float32')
#weight_path = "./test_50000EP/cp.ckpt"
#
#ModelCP = ModelCheckpoint(filepath=weight_path, monitor='loss',
x = np.load('./data/x.npy', allow_pickle=True)
xOBS = np.load('./data/x_obs.npy', allow_pickle=True)
modulus_array = modulus_fit(x[:, 0], x[:, 1])
stiffness_array = stiffness_var(radius, length, thickness, poisson,
modulus_array[:, 0])
damping_array = damping_var(inertia, stiffness_array, modulus_array[:, 0],
modulus_array[:, 1])
stiffness_low = stiffness_array.min()
stiffness_up = stiffness_array.max()
damping_low = damping_array.min()
damping_up = damping_array.max()
input_min = x.min(axis=0)
input_range = x.max(axis=0) - input_min
input_scaling = getScalingDenseLayer(input_min, input_range)
delta_stiffness_mlp = create_mlp(input_scaling, 'delta_stiffness')
delta_damping_mlp = create_mlp(input_scaling, 'delta_damping')
df = pd.read_csv('data/storage_modulus.csv')
storageModulus = arrange_table(df)
df = pd.read_csv('data/loss_modulus.csv')
lossModulus = arrange_table(df)
batch_input_shape = x.shape
select_freq = [0]
select_temp = [1]
# Calculate outputs