morphopt 4.0.0__tar.gz → 4.0.2__tar.gz

{morphopt-4.0.0 → morphopt-4.0.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: morphopt
-Version: 4.0.0
+Version: 4.0.2
 Summary: A morphing optimization framework using PyTorch and torchfea.
 Requires-Python: >=3.12
 Description-Content-Type: text/markdown
@@ -28,14 +28,14 @@ Requires-Dist: pytest>=8.0; extra == "test"
 
 **Developed by Zenan Song**
 
-A differentiable structural optimization framework that eliminates manual sensitivity derivation by fusing the adjoint method with automatic differentiation (AD). The sensitivity corresponds to the virtual work of residual force derivatives on the adjoint displacement field, computed via a single backpropagation through the residual graph. MorphOpt provides a unified interface for geometry, load, and material definition, calls torchfea for GPU-accelerated differentiable nonlinear FEA and adjoint-AD sensitivity, and employs a trust-region optimizer with L-BFGS for design updates. Validated on SIMP topology optimization with B-spline density fields and pneumatic soft robot shape optimization with deformation-dependent follower loads.
+MorphOpt is a differentiable structural optimization framework that combines the adjoint method with automatic differentiation (AD). The sensitivity corresponds to the virtual work of residual force derivatives on the adjoint displacement field, computed via a single backpropagation through the residual graph. MorphOpt provides a unified interface for geometry, load, and material definition, calls torchfea for GPU-accelerated differentiable nonlinear FEA and adjoint-AD sensitivity, and employs a trust-region optimizer with L-BFGS for design updates. Validated on SIMP topology optimization with B-spline density fields and pneumatic soft robot shape optimization with deformation-dependent follower loads.
 
 ## Features
 
 - Custom model definition covering geometry, material, and load parameters
 - Differentiable FEA via torchfea with automatic solving and sensitivity propagation
 - L-BFGS and trust-region methods for parameter updates within a subproblem framework
-- Fully extensible via inheritance — users customize every component in their definition scripts
+- Component customization through framework class inheritance
 
 ## Quick Start
 
@@ -47,7 +47,7 @@ pip install morphopt
 
 Requirements: Python ≥ 3.12; a PyTorch **float64** build is recommended.
 
-> **Optimizations are run as scripts.** Each job is an ordinary Python file that calls `morphopt.start_optimization(...)`. It runs headlessly in a terminal (local or remote), and a finished run can be stopped and resumed later from its result folder. Script execution is the recommended way to run jobs.
+> **Optimizations are run as scripts.** Each job is an ordinary Python file that calls `morphopt.start_optimization(...)`. It runs headlessly in a terminal (local or remote), and optimization runs can be stopped and resumed from their result folders. Script execution is the recommended way to run jobs.
 
 1. **Run a minimal example** — execute it directly in a terminal:
 
@@ -59,7 +59,7 @@ Requirements: Python ≥ 3.12; a PyTorch **float64** build is recommended.
    python examples/gripper.py
    ```
 
-2. **Run your own job** — copy an example as a starting point, adapt the definition (geometry, material, load, updater, objective), and launch it as a script:
+2. **Run a custom job** — copy an example as a starting point, configure the geometry, material, loads, updater, and objective, and launch it as a script:
 
    ```bash
    python my_task.py
@@ -67,20 +67,29 @@ Requirements: Python ≥ 3.12; a PyTorch **float64** build is recommended.
 
    Each run writes its results (logs, FEA files, restart script) into a timestamped result folder under the run root.
 
-3. **Optional — the GUI** is a visual aid for authoring a problem definition and inspecting finished runs in one window. It can also export a runnable script (`.py`), which you can either run directly in the UI or further customize before executing it via Step 2 in a terminal.
+3. **Graphical interface.** The GUI supports problem definition, optimization monitoring, and result inspection in one window. It also exports runnable Python scripts (`.py`) for execution in the UI or a terminal.
 
    ```bash
    morphopt-ui            # console command, available after `pip install morphopt`
-   # python -m morphopt.ui works too (handy when running from a source code directory)
+   # Alternative entry point: python -m morphopt.ui
    ```
 
-The complete UI workflow (definition part + observer part) is described in [`docs/UI_usage.md`](docs/UI_usage.md).
+The GUI opens directly into the full definition workbench with its standard sections present and no problem setup prefilled. Definitions can be created directly in the workbench or generated through **Quickstart**. Quickstart provides shape and topology optimization settings, geometry definitions, and weighted objective terms with associated load conditions. Workspace terms select their driving pressure channels and use boundary integration. Generated problems stay editable and can be saved or exported as runnable scripts.
 
-Further references:
+The [UI quick start](docs/UIguide.md) covers the complete example from model definition to optimization and result inspection. The [script quick start](docs/scriptguide.md) provides the equivalent runnable Python task.
 
-- Task definition guide: `docs/module_definition_guide.md`
-- Module API reference: `docs/module_reference.md`
-- Theory & papers: `docs/theory/`
+### Numeric configuration
+
+Shared UI defaults and tunable kernel parameters live in [`src/morphopt/constants.py`](src/morphopt/constants.py), grouped under `CONSTANT.UI` and `CONSTANT.KERNEL`. Code accesses them directly, for example `CONSTANT.UI.VIEWER.RENDER.LABEL_FONT_SIZE` or `CONSTANT.KERNEL.SOLVER.NUM_PROCESS`. Edit a value there to change its default; restart MorphOpt for the change to take effect. Coordinate indices, tensor dimensions, and coefficients that define mathematical formulas remain in code. UI translations remain in JSON files under `src/morphopt/ui/locales/`.
+
+### Documentation
+
+| Workflow | Quick start | Module and interface reference |
+|---|---|---|
+| Python script | [Start an optimization](docs/scriptguide.md) | [Define and extend each module](docs/scriptreference.md) |
+| Graphical UI | [Create and run the example](docs/UIguide.md) | [Modules, parameters and interface definitions](docs/UIreference.md) |
+
+The [documentation index](docs/README.md) includes the reading order and automatic section numbering commands. Theory and code architecture are documented separately under `docs/theory/` and `docs/codestructure/`.
 
 ## Directory Overview
 
@@ -101,7 +110,7 @@ Research use preferred. For production or commercial use, please evaluate and en
 
 ## Citation
 
-If you use MorphOpt in your research, please cite:
+Research using MorphOpt can cite the following publications:
 
 - TRO 2023
 

{morphopt-4.0.0 → morphopt-4.0.2}/README.md

@@ -2,14 +2,14 @@
 
 **Developed by Zenan Song**
 
-A differentiable structural optimization framework that eliminates manual sensitivity derivation by fusing the adjoint method with automatic differentiation (AD). The sensitivity corresponds to the virtual work of residual force derivatives on the adjoint displacement field, computed via a single backpropagation through the residual graph. MorphOpt provides a unified interface for geometry, load, and material definition, calls torchfea for GPU-accelerated differentiable nonlinear FEA and adjoint-AD sensitivity, and employs a trust-region optimizer with L-BFGS for design updates. Validated on SIMP topology optimization with B-spline density fields and pneumatic soft robot shape optimization with deformation-dependent follower loads.
+MorphOpt is a differentiable structural optimization framework that combines the adjoint method with automatic differentiation (AD). The sensitivity corresponds to the virtual work of residual force derivatives on the adjoint displacement field, computed via a single backpropagation through the residual graph. MorphOpt provides a unified interface for geometry, load, and material definition, calls torchfea for GPU-accelerated differentiable nonlinear FEA and adjoint-AD sensitivity, and employs a trust-region optimizer with L-BFGS for design updates. Validated on SIMP topology optimization with B-spline density fields and pneumatic soft robot shape optimization with deformation-dependent follower loads.
 
 ## Features
 
 - Custom model definition covering geometry, material, and load parameters
 - Differentiable FEA via torchfea with automatic solving and sensitivity propagation
 - L-BFGS and trust-region methods for parameter updates within a subproblem framework
-- Fully extensible via inheritance — users customize every component in their definition scripts
+- Component customization through framework class inheritance
 
 ## Quick Start
 
@@ -21,7 +21,7 @@ pip install morphopt
 
 Requirements: Python ≥ 3.12; a PyTorch **float64** build is recommended.
 
-> **Optimizations are run as scripts.** Each job is an ordinary Python file that calls `morphopt.start_optimization(...)`. It runs headlessly in a terminal (local or remote), and a finished run can be stopped and resumed later from its result folder. Script execution is the recommended way to run jobs.
+> **Optimizations are run as scripts.** Each job is an ordinary Python file that calls `morphopt.start_optimization(...)`. It runs headlessly in a terminal (local or remote), and optimization runs can be stopped and resumed from their result folders. Script execution is the recommended way to run jobs.
 
 1. **Run a minimal example** — execute it directly in a terminal:
 
@@ -33,7 +33,7 @@ Requirements: Python ≥ 3.12; a PyTorch **float64** build is recommended.
    python examples/gripper.py
    ```
 
-2. **Run your own job** — copy an example as a starting point, adapt the definition (geometry, material, load, updater, objective), and launch it as a script:
+2. **Run a custom job** — copy an example as a starting point, configure the geometry, material, loads, updater, and objective, and launch it as a script:
 
    ```bash
    python my_task.py
@@ -41,20 +41,29 @@ Requirements: Python ≥ 3.12; a PyTorch **float64** build is recommended.
 
    Each run writes its results (logs, FEA files, restart script) into a timestamped result folder under the run root.
 
-3. **Optional — the GUI** is a visual aid for authoring a problem definition and inspecting finished runs in one window. It can also export a runnable script (`.py`), which you can either run directly in the UI or further customize before executing it via Step 2 in a terminal.
+3. **Graphical interface.** The GUI supports problem definition, optimization monitoring, and result inspection in one window. It also exports runnable Python scripts (`.py`) for execution in the UI or a terminal.
 
    ```bash
    morphopt-ui            # console command, available after `pip install morphopt`
-   # python -m morphopt.ui works too (handy when running from a source code directory)
+   # Alternative entry point: python -m morphopt.ui
    ```
 
-The complete UI workflow (definition part + observer part) is described in [`docs/UI_usage.md`](docs/UI_usage.md).
+The GUI opens directly into the full definition workbench with its standard sections present and no problem setup prefilled. Definitions can be created directly in the workbench or generated through **Quickstart**. Quickstart provides shape and topology optimization settings, geometry definitions, and weighted objective terms with associated load conditions. Workspace terms select their driving pressure channels and use boundary integration. Generated problems stay editable and can be saved or exported as runnable scripts.
 
-Further references:
+The [UI quick start](docs/UIguide.md) covers the complete example from model definition to optimization and result inspection. The [script quick start](docs/scriptguide.md) provides the equivalent runnable Python task.
 
-- Task definition guide: `docs/module_definition_guide.md`
-- Module API reference: `docs/module_reference.md`
-- Theory & papers: `docs/theory/`
+### Numeric configuration
+
+Shared UI defaults and tunable kernel parameters live in [`src/morphopt/constants.py`](src/morphopt/constants.py), grouped under `CONSTANT.UI` and `CONSTANT.KERNEL`. Code accesses them directly, for example `CONSTANT.UI.VIEWER.RENDER.LABEL_FONT_SIZE` or `CONSTANT.KERNEL.SOLVER.NUM_PROCESS`. Edit a value there to change its default; restart MorphOpt for the change to take effect. Coordinate indices, tensor dimensions, and coefficients that define mathematical formulas remain in code. UI translations remain in JSON files under `src/morphopt/ui/locales/`.
+
+### Documentation
+
+| Workflow | Quick start | Module and interface reference |
+|---|---|---|
+| Python script | [Start an optimization](docs/scriptguide.md) | [Define and extend each module](docs/scriptreference.md) |
+| Graphical UI | [Create and run the example](docs/UIguide.md) | [Modules, parameters and interface definitions](docs/UIreference.md) |
+
+The [documentation index](docs/README.md) includes the reading order and automatic section numbering commands. Theory and code architecture are documented separately under `docs/theory/` and `docs/codestructure/`.
 
 ## Directory Overview
 
@@ -75,7 +84,7 @@ Research use preferred. For production or commercial use, please evaluate and en
 
 ## Citation
 
-If you use MorphOpt in your research, please cite:
+Research using MorphOpt can cite the following publications:
 
 - TRO 2023
 

{morphopt-4.0.0 → morphopt-4.0.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "morphopt"
-version = "4.0.0"
+version = "4.0.2"
 description = "A morphing optimization framework using PyTorch and torchfea."
 readme = "README.md"
 requires-python = ">=3.12"
@@ -39,5 +39,12 @@ testpaths = ["tests"]
 pythonpath = ["src"]
 addopts = "-ra"
 
+[tool.ruff.lint]
+extend-select = ["RUF022"]
+
 [tool.setuptools.package-data]
-"morphopt.ui" = ["templates/*.morph", "templates/*.npz"]
+"morphopt.ui" = [
+    "locales/*.json",
+    "theme/styles/*.qss",
+    "assets/*.svg",
+]

{morphopt-4.0.0 → morphopt-4.0.2}/src/morphopt/__init__.py

@@ -67,21 +67,22 @@ def enable_logging(level=__logging.INFO, log_file=None, file_log_level=__logging
 
 # region modules (loaded after the core classes they subclass)
 from . import codesign, shapeopt, simp
+from .codesign import OffsetBoundaryPart, UpdaterOffsetBoundaryPart
 from .opt_runner import debug_optimization, start_optimization
 from .optcore.controller import Controller
 from .optcore.history import History
 from .optcore.modelparams import (
-    BasePartInterface,
+    BaseModelComponent,
+    BasePart,
+    BoxPart,
+    CADPart,
     FEAParams,
     GeometryParams,
-    INPPartInterface,
+    INPPart,
     MaterialsParams,
     Params,
-    TorchFEAPartInterface,
-    load_model_assembly,
-    resolve_model_path,
 )
-from .optcore.modelparams.materialinterface import HomogeneousMaterial
+from .optcore.modelparams.materials import HomogeneousMaterial
 from .optcore.objfunc import ObjectiveFunction
 from .optcore.protocal import (
     ProtocalInitializable,
@@ -91,7 +92,14 @@ from .optcore.protocal import (
 )
 from .optcore.solver import Solver
 from .optcore.updaters import Updaters
-from .shapeopt import BoundaryPartInterface, UpdaterBoundaryPart
+from .shapeopt import (
+    BaseSurfaceConstraints,
+    BoundaryPart,
+    BspSurfaceRelationConstraint,
+    CPGEOSurfaceRelationConstraint,
+    SurfaceRelationConstraint,
+    UpdaterBoundaryPart,
+)
 from .simp import SIMP_BSPFieldMaterials, UpdaterSIMPMaterial
 
 # endregion
@@ -101,3 +109,43 @@ from .utils import check_gradients, get_controller
 # endregion
 
 controller: Controller = None
+
+
+__all__ = [
+    "BaseModelComponent",
+    "BasePart",
+    "BaseSurfaceConstraints",
+    "BoundaryPart",
+    "BoxPart",
+    "BspSurfaceRelationConstraint",
+    "CADPart",
+    "CPGEOSurfaceRelationConstraint",
+    "Controller",
+    "FEAParams",
+    "GeometryParams",
+    "History",
+    "HomogeneousMaterial",
+    "INPPart",
+    "MaterialsParams",
+    "ObjectiveFunction",
+    "OffsetBoundaryPart",
+    "Params",
+    "ProtocalInitializable",
+    "ProtocalSavable",
+    "ProtocalUpdatable",
+    "ProtocalVisualizable",
+    "SIMP_BSPFieldMaterials",
+    "Solver",
+    "SurfaceRelationConstraint",
+    "UpdaterBoundaryPart",
+    "UpdaterOffsetBoundaryPart",
+    "UpdaterSIMPMaterial",
+    "Updaters",
+    "check_gradients",
+    "codesign",
+    "debug_optimization",
+    "get_controller",
+    "shapeopt",
+    "simp",
+    "start_optimization",
+]

morphopt-4.0.2/src/morphopt/codesign/__init__.py

@@ -0,0 +1,16 @@
+"""Cavity offset Parts and their geometry updater.
+
+Compose parameters, objectives, solvers and updater collections using the
+common morphopt interfaces. SIMP materials use the simp interfaces.
+"""
+
+from .offset_boundary_part import OffsetBoundaryPart
+from .update_offset_boundary_part import UpdaterOffsetBoundaryPart
+from .constraints import InwardCurvatureRadius, OffsetSurfaceMinThickness
+
+__all__ = [
+    "OffsetBoundaryPart",
+    "UpdaterOffsetBoundaryPart",
+    "InwardCurvatureRadius",
+    "OffsetSurfaceMinThickness",
+]

{morphopt-4.0.0 → morphopt-4.0.2}/src/morphopt/codesign/constraints.py

@@ -1,7 +1,8 @@
 import torch
 
+from ..constants import CONSTANT
 from ..shapeopt.objectivefuncs.basefuncs import BaseConstraints
-from .geometry import CodesignBoundaryPartInterface
+from .offset_boundary_part import OffsetBoundaryPart
 
 
 class InwardCurvatureRadius(BaseConstraints):
@@ -17,13 +18,13 @@ class InwardCurvatureRadius(BaseConstraints):
         self,
         surface_start: int = 1,
         surface_ids: list[int] | None = None,
-        margin: float = 0.4,
-        margin_ratio: float = 0.45,
-        barrier_thre_ratio: float = 0.1,
-        barrier_ratio: float = 0.0,
-        p: int = 8,
-        penalty_scale: float = 1e3,
-        hard_violation_beta: float = 8.0,
+        margin: float = CONSTANT.KERNEL.CODESIGN.CURVATURE_MARGIN,
+        margin_ratio: float = CONSTANT.KERNEL.CODESIGN.CURVATURE_MARGIN_RATIO,
+        barrier_thre_ratio: float = CONSTANT.KERNEL.CODESIGN.CURVATURE_BARRIER_THRESHOLD_RATIO,
+        barrier_ratio: float = CONSTANT.KERNEL.CODESIGN.BARRIER_PENALTY_RATIO,
+        p: int = CONSTANT.KERNEL.CODESIGN.BARRIER_EXPONENT,
+        penalty_scale: float = CONSTANT.KERNEL.CODESIGN.PENALTY_SCALE,
+        hard_violation_beta: float = CONSTANT.KERNEL.CODESIGN.HARD_VIOLATION_BETA,
         normal_sign: float = 1.0,
     ) -> None:
         """
@@ -71,7 +72,9 @@ class InwardCurvatureRadius(BaseConstraints):
 
     @staticmethod
     def _principal_curvatures(
-        rdu: torch.Tensor, rdu2: torch.Tensor, eps: float = 1e-12
+        rdu: torch.Tensor,
+        rdu2: torch.Tensor,
+        eps: float = CONSTANT.KERNEL.CODESIGN.GEOMETRY_TOLERANCE,
     ) -> tuple[torch.Tensor, torch.Tensor]:
         normal0 = torch.cross(rdu[:, :, 1], rdu[:, :, 0], dim=1)
         normal = normal0 / (normal0.norm(dim=1, keepdim=True) + eps)
@@ -96,11 +99,11 @@ class InwardCurvatureRadius(BaseConstraints):
         """Take the Part this constraint is evaluated for.
 
         The geometry updater hands over the Part it evaluates; only a
-        :class:`~morphopt.codesign.CodesignBoundaryPartInterface` carries a
+        :class:`~morphopt.codesign.OffsetBoundaryPart` carries a
         shell, so anything else is rejected here.
         """
         super().initialize(r0=r0, rdu0=rdu0, *args, **kwargs)
-        if not isinstance(part_interface, CodesignBoundaryPartInterface):
+        if not isinstance(part_interface, OffsetBoundaryPart):
             raise TypeError(
                 f"{type(self).__name__} is evaluated for one co-design Part "
                 f"(got {type(part_interface).__name__})."
@@ -127,9 +130,15 @@ class InwardCurvatureRadius(BaseConstraints):
             if t <= 0:
                 continue
 
-            margin_abs = max(float((self.margin_ratio * t) + self.margin), 1e-12)
+            margin_abs = max(
+                float((self.margin_ratio * t) + self.margin),
+                CONSTANT.KERNEL.CODESIGN.GEOMETRY_TOLERANCE,
+            )
             k_limit = 1.0 / (t + margin_abs)
-            barrier_thre = max(k_limit * self.barrier_thre_ratio, 1e-12)
+            barrier_thre = max(
+                k_limit * self.barrier_thre_ratio,
+                CONSTANT.KERNEL.CODESIGN.GEOMETRY_TOLERANCE,
+            )
 
             k1, k2 = self._principal_curvatures(rdu[i], rdu2[i])
 
@@ -146,7 +155,8 @@ class InwardCurvatureRadius(BaseConstraints):
                 rel_violation = (violation[index] / barrier_thre).clamp(min=0.0)
                 hard_boost = torch.exp(
                     (self.hard_violation_beta * (rel_violation - 1.0)).clamp(
-                        min=0.0, max=20.0
+                        min=0.0,
+                        max=CONSTANT.KERNEL.CODESIGN.HARD_VIOLATION_EXPONENT_MAX,
                     )
                 )
                 penalty_strong = self.penalty_scale * penalty * hard_boost
@@ -168,16 +178,16 @@ class OffsetSurfaceMinThickness(BaseConstraints):
         surface_start: int = 1,
         surface_ids: list[int] | None = None,
         min_distance: float | None = None,
-        search_ratio: float = 1.5,
-        barrier_thre_ratio: float = 0.05,
-        barrier_ratio: float = 0.0,
-        p: int = 8,
-        penalty_scale: float = 1e3,
-        hard_violation_beta: float = 8.0,
+        search_ratio: float = CONSTANT.KERNEL.CODESIGN.THICKNESS_SEARCH_RATIO,
+        barrier_thre_ratio: float = CONSTANT.KERNEL.CODESIGN.THICKNESS_BARRIER_THRESHOLD_RATIO,
+        barrier_ratio: float = CONSTANT.KERNEL.CODESIGN.BARRIER_PENALTY_RATIO,
+        p: int = CONSTANT.KERNEL.CODESIGN.BARRIER_EXPONENT,
+        penalty_scale: float = CONSTANT.KERNEL.CODESIGN.PENALTY_SCALE,
+        hard_violation_beta: float = CONSTANT.KERNEL.CODESIGN.HARD_VIOLATION_BETA,
         normal_sign: float = 1.0,
-        normal_opposition_thre: float = 0.0,
-        normal_min: float = -0.7,
-        normal_max: float = -0.5,
+        normal_opposition_thre: float = CONSTANT.KERNEL.CODESIGN.NORMAL_OPPOSITION_THRESHOLD,
+        normal_min: float = CONSTANT.KERNEL.CODESIGN.NORMAL_TRANSITION_MIN,
+        normal_max: float = CONSTANT.KERNEL.CODESIGN.NORMAL_TRANSITION_MAX,
     ) -> None:
         """
         Initialize the minimum thickness constraint for the offset surface.
@@ -233,7 +243,10 @@ class OffsetSurfaceMinThickness(BaseConstraints):
     @staticmethod
     def _surface_normals(rdu_sf: torch.Tensor) -> torch.Tensor:
         normal0 = torch.cross(rdu_sf[:, :, 1], rdu_sf[:, :, 0], dim=1)
-        normals = normal0 / (normal0.norm(dim=1, keepdim=True) + 1e-12)
+        normals = normal0 / (
+            normal0.norm(dim=1, keepdim=True)
+            + CONSTANT.KERNEL.CODESIGN.GEOMETRY_TOLERANCE
+        )
         return normals
 
     def _offset_points(
@@ -249,17 +262,17 @@ class OffsetSurfaceMinThickness(BaseConstraints):
             if self.min_distance is not None
             else (2.0 * thickness)
         )
-        return max(dmin, 1e-12)
+        return max(dmin, CONSTANT.KERNEL.CODESIGN.GEOMETRY_TOLERANCE)
 
     def initialize(self, r0, rdu0, *args, part_interface=None, **kwargs) -> None:
         """Take the Part this constraint is evaluated for and pair its points.
 
         The geometry updater hands over the Part it evaluates; only a
-        :class:`~morphopt.codesign.CodesignBoundaryPartInterface` carries a
+        :class:`~morphopt.codesign.OffsetBoundaryPart` carries a
         shell, so anything else is rejected here.
         """
         super().initialize(r0=r0, rdu0=rdu0, *args, **kwargs)
-        if not isinstance(part_interface, CodesignBoundaryPartInterface):
+        if not isinstance(part_interface, OffsetBoundaryPart):
             raise TypeError(
                 f"{type(self).__name__} is evaluated for one co-design Part "
                 f"(got {type(part_interface).__name__})."
@@ -318,7 +331,8 @@ class OffsetSurfaceMinThickness(BaseConstraints):
             if thickness <= 0:
                 continue
             barrier_thre = max(
-                self._min_distance(thickness) * self.barrier_thre_ratio, 1e-12
+                self._min_distance(thickness) * self.barrier_thre_ratio,
+                CONSTANT.KERNEL.CODESIGN.GEOMETRY_TOLERANCE,
             )
 
             # Base surface normals (offset surface shares same normal direction)
@@ -343,7 +357,10 @@ class OffsetSurfaceMinThickness(BaseConstraints):
                 r_sf=r[sf_idx], rdu_sf=rdu[sf_idx], thickness=thickness
             )
             delta = roff[pairs_n[0]] - roff[pairs_n[1]]
-            dist = torch.sqrt((delta * delta).sum(dim=1) + 1e-18)
+            dist = torch.sqrt(
+                (delta * delta).sum(dim=1)
+                + CONSTANT.KERNEL.CODESIGN.DISTANCE_SQUARED_TOLERANCE
+            )
 
             violation = mindist_n - dist + barrier_thre
             indexl, l = self.barrier_function(
@@ -361,7 +378,8 @@ class OffsetSurfaceMinThickness(BaseConstraints):
                 rel_violation = (violation[indexl] / barrier_thre).clamp(min=0.0)
                 hard_boost = torch.exp(
                     (self.hard_violation_beta * (rel_violation - 1.0)).clamp(
-                        min=0.0, max=20.0
+                        min=0.0,
+                        max=CONSTANT.KERNEL.CODESIGN.HARD_VIOLATION_EXPONENT_MAX,
                     )
                 )
                 penalty_strong = self.penalty_scale * l * hard_boost * np_smooth

morphopt-4.0.0/src/morphopt/codesign/geometry.py → morphopt-4.0.2/src/morphopt/codesign/offset_boundary_part.py

@@ -1,4 +1,4 @@
-"""Co-design boundary part interface: cavity surfaces with an offset shell.
+"""Co-design boundary part: cavity surfaces with an offset shell.
 
 The co-design (``codesign``) scheme extends the shape-optimization boundary
 description with a solid shell: every cavity surface (index ``1..n``) is offset
@@ -13,12 +13,13 @@ import numpy as np
 import torch
 import torchfea
 
-from ..shapeopt.boundarypartinterface import BoundaryPartInterface
+from ..constants import CONSTANT
+from ..shapeopt.boundary_part import BoundaryPart
 
-__all__ = ["CodesignBoundaryPartInterface"]
+__all__ = ["OffsetBoundaryPart"]
 
 
-class CodesignBoundaryPartInterface(BoundaryPartInterface):
+class OffsetBoundaryPart(BoundaryPart):
     """Boundary-surface geometry whose cavity surfaces carry a solid shell.
 
     Parameters
@@ -31,7 +32,7 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
 
     def __init__(
         self,
-        shell_thickness: float = 1.0,
+        shell_thickness: float = CONSTANT.KERNEL.CODESIGN_GEOMETRY.DEFAULT_SHELL_THICKNESS,
         num_layers: int = 1,
         **kwargs: object,
     ) -> None:
@@ -40,17 +41,16 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
         self.shell_thickness: float = float(shell_thickness)
         """Total thickness of the offset shell on the cavity surfaces."""
 
-        self.num_layers: int = max(1, int(num_layers))
+        self.num_layers: int = int(num_layers)
         """Number of wedge layers through the shell thickness."""
 
         self._node_idx: list[np.ndarray] = []
         """Node indices of each offset layer (0 = base surface mesh)."""
 
-        self._tri_local: np.ndarray | None = None
-        """Cavity triangle connectivity stored with local node indices."""
-
-        self._shell_elem_offset: int | None = None
-        """Element-id offset of the shell block (0-based)."""
+        if self.shell_thickness <= 0:
+            raise ValueError("shell_thickness must be positive.")
+        if self.num_layers < 1:
+            raise ValueError("num_layers must be at least one.")
 
     # ------------------------------------------------------------- shell mesh
     def _compute_normals(self, base_nodes: torch.Tensor) -> torch.Tensor:
@@ -60,8 +60,9 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
             surface = self.surfaceinterfaces[index]
             normals.append(
                 surface.get_normals(
-                    torch.from_numpy(surface.surf_node_uv).to(
-                        torch.get_default_device()
+                    torch.as_tensor(
+                        surface.surf_node_uv, device=base_nodes.device,
+                        dtype=base_nodes.dtype,
                     )
                 )
             )
@@ -76,16 +77,19 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
         """Add the offset shell wedge elements to the meshed Part."""
         return self._build_shell_c3d6(part)
 
-    def _build_shell_c3d6(self, part: torchfea.Part) -> torchfea.Part:
-        thickness = float(self.shell_thickness)
-        if thickness <= 0 or self.num_layers <= 0:
-            return part
+    def build_part(self, path_result=None, pools=None) -> torchfea.Part:
+        """Build and validate the offset wedges, including quadratic conversion."""
+        part = super().build_part(path_result=path_result, pools=pools)
+        shell = part.elems["C3D6"]
+        shell.initialize(nodes=part.nodes)
+        weights = shell.gaussian_weight
+        if not torch.isfinite(weights).all() or (weights <= 0).any():
+            raise ValueError(f"Invalid offset shell mesh on Part {self.part_name!r}.")
+        return part
 
+    def _build_shell_c3d6(self, part: torchfea.Part) -> torchfea.Part:
         # Ensure surface elements exist so we can query triangle meshes.
-        try:
-            part.surfaces.initialize(part)
-        except Exception:
-            pass
+        part.surfaces.initialize(part)
 
         tri_by_surface: dict[int, np.ndarray] = {}
         tri_local_by_surface: dict[int, np.ndarray] = {}
@@ -100,7 +104,7 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
             cavity_order.append(sidx)
             tri_by_surface[sidx] = tri
             base_ids_blocks.append(
-                np.sort(np.asarray(part.set_nodes[surf_name], dtype=int))
+                np.asarray(self.surfaceinterfaces[sidx].surf_node_idx, dtype=int)
             )
 
         if len(tri_by_surface) == 0:
@@ -115,7 +119,6 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
         local_map = -np.ones((int(part.nodes.shape[0]),), dtype=int)
         local_map[base_ids] = np.arange(base_ids.shape[0], dtype=int)
 
-        tri_local_all = []
         for sidx in cavity_order:
             tri_global = tri_by_surface[sidx]
             tri_local = local_map[tri_global]
@@ -125,8 +128,6 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
                 )
             tri_local = tri_local.astype(int)
             tri_local_by_surface[sidx] = tri_local
-            tri_local_all.append(tri_local)
-        self._tri_local = np.concatenate(tri_local_all, axis=0).astype(int)
 
         device = part.nodes.device
         base_ids_t = torch.as_tensor(base_ids, dtype=torch.long, device=device)
@@ -155,7 +156,6 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
                 continue
             existing_max = max(existing_max, int(element._elems_index.max().item()))
         elem_offset = int(existing_max + 1)
-        self._shell_elem_offset = elem_offset
 
         # Build linear wedge connectivity (C3D6) for all layers.
         wedge6_blocks = []
@@ -217,10 +217,7 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
                 )
 
         # Refresh the surfaces cache to include the newly added offset surfaces.
-        try:
-            part.surfaces.initialize(part)
-        except Exception:
-            pass
+        part.surfaces.initialize(part)
 
         return part
 
@@ -231,9 +228,6 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
         """Update the Part nodes and keep the offset layers consistent."""
         super().modify_assembly(design_sensitivity_vars, assembly)
 
-        if len(self._node_idx) == 0 or self._tri_local is None:
-            return
-
         part = assembly.get_part(self.part_name)
         nodes = part.nodes.clone()
 
@@ -262,4 +256,8 @@ class CodesignBoundaryPartInterface(BoundaryPartInterface):
     @staticmethod
     def _default_opacity(index: int) -> float:
         """The solid volume is hidden so the shell stays visible."""
-        return 0.0 if index == 0 else 1.0
+        return (
+            0.0
+            if index == 0
+            else CONSTANT.KERNEL.VISUALIZATION.OFFSET_SHELL_SURFACE_OPACITY
+        )