hook design

Author: HYB777

model-based-design/hook_design/IPOPT.out

@@ -0,0 +1,59 @@
+# Hook Design
+
+The objective in the hook design is to identify an optimal shape that enhances the stiffness
+while adhering to specific volume constraints and loads. This can be achieved by minimizing
+the strain energy.
+
+## Shape Sampling
+
+We model the half of the hook by lofting 11 2D sections along a guideline, see Appendix G for more details.
+For maintaining a naturalistic semblance characterized by smoothness and minimal undulation, 
+we apply mean filter to each section once or two at random (each section are generated independently and randomly),
+considering its neighboring sections.
+
+Finally, we generate a total of 222,141 shape data points. All data are stored in [data](data).
+
+## High-fidelity Physical Model Learning
+
+We employ the mean-teacher-based active learning algorithm on this dataset, to select 50,000 shapes
+for labeling and train the physical model, the corresponding labels are obtained by COMSOL.
+
+Run the following code under the directory [mean-teacher-al](mean-teacher-al).
+
+```
+python main.py
+```
+
+## Shape Anomaly Detection
+
+Firstly, run the following code to compute the intrinsic dimension.
+
+```
+python lpca.py
+```
+
+The results should be 68 (round up). We choose 68 as the dimension of the latent space in
+the auto-encoder.
+
+Run the following code for training.
+
+```
+python main_ae.py --embed-dim 68
+```
+
+The trained model are stored in [results_aeID=68](shape-anomaly-detection/results_aeID=68).
+
+## Numerical Optimization
+
+We perform two kinds of optimization problem.
+
+* P1: $\min E$ s.t. $V\leq0.5V_{\text{init}}$
+* P2: $\min V$ s.t. $E\leq0.5E_{\text{init}}$
+
+Run the following code
+
+```
+python shape_opt_pysparse.py --prob minEconV # problem 1
+
+python shape_opt_pysparse.py --prob minVconE # problem 2
+```

model-based-design/hook_design/README.md

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0fe9b642359e81f26d0974966a5daef0ea3e3cb0e2bbea9b79baf33c56c9cd3
+size 12665079

model-based-design/hook_design/data/data_test.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c3db8073bc033f72c3c2550dd8515901545f234b716784ac7fb97628ce02741
+size 101319387

model-based-design/hook_design/data/data_train.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0928d38368881c6e3df45ce06e3ba0d57b887560b84816b8c827c74b61b15d49
+size 12665079

model-based-design/hook_design/data/data_val.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f34dcac26f1ecfca0be8305efc1db689a4d3de49a6b2b44316e82786b446898
+size 126648667

model-based-design/hook_design/data/params_dataset.npy

@@ -0,0 +1,56 @@
+import numpy as np
+import torch
+import scipy.io as scio
+from torch.utils.data.dataset import Dataset
+
+
+def mat2npy(mat_str_list):
+    mat_list = [scio.loadmat(it) for it in mat_str_list]
+    scales = np.vstack([it['scales'] if 'scales' in it.keys() else it['scales_i'] for it in mat_list])
+    sections_pts = np.vstack([it['section_pts'].reshape(-1, 5*10) if 'section_pts' in it.keys() else it['section_pts_i'].reshape(-1, 5*10) for it in mat_list])
+    guide_pts = np.vstack([it['guide_pts'] if 'guide_pts' in it.keys() else it['guide_pts_i'] for it in mat_list])
+    labels = np.vstack([it['results'] if 'results' in it.keys() else it['results_i'] for it in mat_list])
+    params = np.hstack([scales, sections_pts, guide_pts])
+    return params, labels
+
+
+class DataLoaderFit(Dataset):
+    def __init__(self, params, labels):
+        """
+        :param params: (N, 11 + 11*5 + 8) [scales, sections_pts, guide_pts]
+        :param labels: (N, 2)
+        """
+        super(DataLoaderFit, self).__init__()
+
+        self.data = torch.from_numpy(params).float()
+        self.labels = torch.from_numpy(labels).float()
+
+    def __len__(self):
+        return len(self.labels)
+
+    def __getitem__(self, item):
+        data = self.data[item]
+        label = self.labels[item]
+
+        return data, label
+
+
+class DataLoaderAE(Dataset):
+    def __init__(self, params):
+        """
+        :param params: (N, 11 + 11*5 + 8) [scales, sections_pts, guide_pts]
+        """
+        super(DataLoaderAE, self).__init__()
+
+        self.data = torch.from_numpy(params).float()
+
+    def __len__(self):
+        return len(self.data)
+
+    def __getitem__(self, item):
+
+        data_tensor = self.data[item]
+
+        return data_tensor
+
+