| from __future__ import annotations | |
| import torch | |
| from src import train_t2_physics_integrator as subject | |
| def parameters() -> torch.Tensor: | |
| return torch.tensor( | |
| [[190.0e9, 0.3, 280.0e6, 0.0, 35.0e9, 60.0, 8.0e9, 8.0, 70.0e6, 8.0]], | |
| dtype=torch.float32, | |
| ) | |
| def test_physics_integrator_preserves_hard_state_constraints() -> None: | |
| strain = torch.zeros((1, 31, 6), dtype=torch.float32) | |
| strain[0, :, 0] = torch.linspace(0.0, 0.007, 31) | |
| strain[0, :, 1] = -0.5 * strain[0, :, 0] | |
| strain[0, :, 2] = -0.5 * strain[0, :, 0] | |
| model = subject.PlasticIncrementNet() | |
| mean = torch.zeros(10) | |
| std = torch.ones(10) | |
| result = subject.rollout(strain, parameters(), model, mean, std) | |
| assert torch.all(result["peeq"][:, 1:] >= result["peeq"][:, :-1]) | |
| assert float(result["plastic_strain"][..., :3].sum(dim=-1).abs().max().detach()) < 1.0e-6 | |
| assert torch.isfinite(result["stress"]).all() | |
| def test_elastic_path_remains_elastic() -> None: | |
| strain = torch.zeros((1, 11, 6), dtype=torch.float32) | |
| strain[0, :, 0] = torch.linspace(0.0, 1.0e-5, 11) | |
| model = subject.PlasticIncrementNet() | |
| result = subject.rollout(strain, parameters(), model, torch.zeros(10), torch.ones(10)) | |
| assert torch.count_nonzero(result["plastic_increment"]) == 0 | |
| assert torch.count_nonzero(result["peeq"]) == 0 | |
| def test_neural_correction_is_bounded() -> None: | |
| model = subject.PlasticIncrementNet() | |
| values = model(torch.randn(50, 17)) | |
| assert float(values.abs().max()) <= 0.750001 | |