def cracked_concrete_plots(c: Config):
"""Contour plots of cracked concrete scenarios."""
response_type = ResponseType.YTranslation
# 10 x 10 grid of points on the bridge deck where to record fem.
points = [
Point(x=x, y=0, z=z) for x, z in itertools.product(
np.linspace(c.bridge.x_min, c.bridge.x_max, 10),
np.linspace(c.bridge.z_min, c.bridge.z_max, 10),
)
]
# Create empty traffic array and collect fem.
response_array = responses_to_traffic_array(
c=c,
traffic_array=load_normal_traffic_array(c)[0],
response_type=response_type,
bridge_scenario=cracked_scenario,
points=points,
sim_runner=OSRunner,
)
for t in range(len(response_array)):
top_view_bridge(c.bridge, abutments=True, piers=True)
responses = Responses.from_responses(
response_type=response_type,
responses=[(response_array[t][p], point)
for p, point in enumerate(points)],
)
plot_contour_deck(c=c, responses=responses, center_norm=True)
plt.title("Cracked Concrete")
plt.savefig(c.get_image_path("cracked-scenario", f"cracked-time-{t}"))
plt.close()
def pairwise_cluster(c: Config, load: bool):
"""Cluster pairwise maps from healthy and damaged scenarios."""
features_path = c.get_data_path("features", "pairwise-cluster", bridge=False)
if not load:
normal_traffic_array, _ = load_normal_traffic_array(c=c, mins=24)
normal_traffic_array = normal_traffic_array[
int(len(normal_traffic_array) / 24) :
]
response_type = ResponseType.YTranslation
grid_points = [
Point(x=x, y=0, z=-9.65)
for x, _ in itertools.product(
np.linspace(c.bridge.x_min, c.bridge.x_max, 50),
# np.linspace(c.bridge.x_min, c.bridge.x_max, 4),
[1],
)
]
# Collect a list of features per scenarios scenario.
features = []
for damage_scenario in healthy_and_cracked_scenarios[1:]:
damage_c = damage_scenario.use(c)
responses = responses_to_traffic_array(
c=damage_c,
traffic_array=normal_traffic_array,
response_type=response_type,
bridge_scenario=damage_scenario,
points=grid_points,
sim_runner=OSRunner,
).T
ks_values = []
for p0_i, point0 in enumerate(grid_points):
print_i(f"Point {p0_i + 1} / {len(grid_points)}", end="\r")
ks_values.append([])
for p1_i, point1 in enumerate(grid_points):
ks = ks_no_outliers(responses[p0_i], responses[p1_i])
ks_values[-1].append(ks)
features.append((ks_values, damage_scenario.name))
# Save features to disk.
features = np.array(features)
np.save(features_path, features)
features = np.load(features_path)
# Reduce each pairwise map to a sum per sensor.
for f_i, (feature, feature_name) in enumerate(features):
features[f_i] = ([sum(sensor) for sensor in feature], feature_name)
features[f_i] = ([sum(sensor) for sensor in features[f_i]], feature_name)
# Cluster each pairwise map.
from sklearn.cluster import KMeans
kmeans = KMeans(n_clusters=2)
kmeans.fit(features)
def traffic_response_plots(c: Config, times: int = 3):
"""Response to normal traffic per scenarios scenario at multiple time steps."""
response_type = ResponseType.YTranslation
# 10 x 10 grid of points on the bridge deck where to record fem.
points = [
Point(x=x, y=0, z=z) for x, z in itertools.product(
np.linspace(c.bridge.x_min, c.bridge.x_max, 10),
np.linspace(c.bridge.z_min, c.bridge.z_max, 10),
)
]
# for damage_scenario in all_scenarios(c):
for damage_scenario in [unit_temp_scenario]:
response_array = responses_to_traffic_array(
c=c,
traffic_array=load_normal_traffic_array(c, mins=1)[0],
response_type=response_type,
bridge_scenario=damage_scenario,
points=points,
sim_runner=OSRunner,
)
print(response_array.shape)
mean_response_array = np.mean(response_array, axis=0).T
print(mean_response_array.shape)
print(mean_response_array.shape)
for t in range(times):
time_index = -1 + abs(t)
top_view_bridge(c.bridge, abutments=True, piers=True)
responses = Responses.from_responses(
response_type=response_type,
responses=[(response_array[time_index][p], point)
for p, point in enumerate(points)],
)
plot_contour_deck(c=c,
responses=responses,
center_norm=True,
levels=100)
plt.title(damage_scenario.name)
plt.savefig(
c.get_image_path(
"contour-traffic-response",
f"{damage_scenario.name}-time={time_index}",
))
plt.close()
def oneclass(c: Config):
normal_traffic_array, traffic_scenario = load_normal_traffic_array(c)
bridge_scenarios = [HealthyScenario()] + each_pier_scenarios(c)
response_type = ResponseType.YTranslation
points = [
Point(x=x, y=0, z=z)
for x, z in itertools.product(
np.linspace(c.bridge.x_min, c.bridge.x_max / 2, 20),
np.linspace(c.bridge.z_min, c.bridge.z_max / 2, 3),
)
]
results = []
for b, bridge_scenario in enumerate(bridge_scenarios):
print_i(f"One class: bridge scenario {bridge_scenario.name}")
responses = responses_to_traffic_array(
c=c,
traffic_array=normal_traffic_array,
response_type=response_type,
bridge_scenario=bridge_scenario,
points=points,
fem_runner=OSRunner(c),
).T
print(len(normal_traffic_array))
print(responses.shape)
# Fit on the healthy scenario.
if b == 0:
assert len(responses) == len(points)
clfs = []
for r, rs in enumerate(responses):
print_i(f"Training classifier {r} / {len(responses)}")
clfs.append(OneClassSVM().fit(rs.reshape(-1, 1)))
scenario_results = []
for p, _ in enumerate(points):
print_i(f"Predicting points {p} / {len(points)}")
prediction = clfs[p].predict(responses[p].reshape(-1, 1))
print(prediction)
print(len(prediction[prediction < 0]))
print(len(prediction[prediction > 0]))
def pairwise_sensors(c: Config, dist_measure=ks_no_outliers):
"""Compare distribution of pairs of sensors under HealthyScenario."""
normal_traffic_array, traffic_scenario = load_normal_traffic_array(c)
response_type = ResponseType.YTranslation
points = [
Point(x=x, y=0, z=z)
for x, z in itertools.product(
np.linspace(c.bridge.x_min, c.bridge.x_max, 50),
np.linspace(c.bridge.z_min, c.bridge.z_max, 4),
)
]
bridge_scenario = HealthyScenario()
responses = responses_to_traffic_array(
c=c,
traffic_array=normal_traffic_array,
response_type=response_type,
bridge_scenario=bridge_scenario,
points=points,
sim_runner=OSRunner,
).T
assert len(responses) == len(points)
ks_values_healthy = []
for p0, point0 in enumerate(points):
print_i(f"Point {p0 + 1} / {len(points)}")
ks_values_healthy.append([])
for p1, point1 in enumerate(points):
ks = dist_measure(responses[p0], responses[p1])
ks_values_healthy[-1].append(ks)
plt.landscape()
plt.imshow(ks_values_healthy)
plt.savefig(c.get_image_path("joint-clustering", "healthy-bridge"))
plt.close()
bridge_scenario = each_pier_scenarios(c)[0]
responses = responses_to_traffic_array(
c=c,
traffic_array=normal_traffic_array,
response_type=response_type,
bridge_scenario=bridge_scenario,
points=points,
sim_runner=OSRunner,
).T
assert len(responses) == len(points)
ks_values_damage = []
for p0, point0 in enumerate(points):
print_i(f"Point {p0 + 1} / {len(points)}")
ks_values_damage.append([])
for p1, point1 in enumerate(points):
ks = dist_measure(responses[p0], responses[p1])
ks_values_damage[-1].append(ks)
plt.imshow(ks_values_damage)
plt.savefig(c.get_image_path("joint-clustering", "scenarios-bridge"))
plt.close()
ks_values_comp = []
for p0, point0 in enumerate(points):
ks_values_comp.append([])
for p1, point1 in enumerate(points):
comp = abs(ks_values_healthy[p0][p1] - ks_values_damage[p0][p1])
ks_values_comp[-1].append(comp)
plt.landscape()
plt.imshow(ks_values_comp)
plt.savefig(c.get_image_path("joint-clustering", "scenarios-bridge-comp"))
plt.close()
responses = Responses.from_responses(
response_type=response_type,
responses=[(sum(ks_values_comp[p]), point) for p, point in enumerate(points)],
)
top_view_bridge(c.bridge, abutments=True, piers=True)
plot_contour_deck(c=c, responses=responses)
plt.savefig(c.get_image_path("joint-clustering", "scenarios-bridge-comp-contour"))
plt.close()