code/t4_structural_dynamics_v1.py

"""Resumable T4 v1 structural-dynamics campaign built on the validated pilot."""

from __future__ import annotations

import argparse
from copy import deepcopy
import json
import os
from pathlib import Path
import shutil
import tempfile
from time import perf_counter

import h5py
import numpy as np
from scipy.stats import qmc

try:
    from .t4_structural_dynamics import (
        ROOT,
        analytical_first_frequency,
        read_field_history,
        run_trajectory,
        solve_modes,
    )
except ImportError:  # Direct script execution.
    from t4_structural_dynamics import (
        ROOT,
        analytical_first_frequency,
        read_field_history,
        run_trajectory,
        solve_modes,
    )


CONFIG_PATH = ROOT / "configs" / "t4_structural_dynamics_v1.json"
DATA_DIR = ROOT / "data" / "t4_structural_dynamics_v1"
CASE_DIR = DATA_DIR / "cases"
REPRESENTATIVE_DIR = DATA_DIR / "representative_fields"
PARAMETERS = (
    "length_m",
    "height_m",
    "young_pa",
    "density_kg_m3",
    "damping_ratio",
    "traction_amplitude_pa",
)


def load_config(path: Path = CONFIG_PATH) -> dict:
    return json.loads(path.read_text(encoding="utf-8"))


def _map_ranges(unit: np.ndarray, ranges: dict) -> dict[str, float]:
    return {
        name: float(ranges[name][0] + unit[index] * (ranges[name][1] - ranges[name][0]))
        for index, name in enumerate(PARAMETERS)
    }


def configuration_design(config: dict | None = None) -> list[dict]:
    selected = load_config() if config is None else config
    count = int(selected["configuration_count"])
    if count != 128:
        raise ValueError("The frozen v1 Sobol design requires 128 configurations.")
    unit = qmc.Sobol(d=len(PARAMETERS), scramble=True, seed=int(selected["seed"])).random_base2(7)
    interior = selected["interior_parameter_ranges"]
    full = selected["full_parameter_ranges"]
    rows: list[dict] = []
    for index in range(count):
        parameters = _map_ranges(unit[index], interior)
        if index < 96:
            split, role = "train", "id"
        elif index < 112:
            split, role = "validation", "id"
        elif index < 120:
            split, role = "test", "id"
        elif index < 124:
            split, role = "test", "excitation_ood"
        else:
            split, role = "test", "parameter_ood"

        # The four parameter-OOD cases jointly cover all six physical inputs.
        if index == 124:
            parameters["length_m"] = full["length_m"][0]
            parameters["damping_ratio"] = full["damping_ratio"][1]
        elif index == 125:
            parameters["height_m"] = full["height_m"][0]
            parameters["young_pa"] = full["young_pa"][0]
        elif index == 126:
            parameters["length_m"] = full["length_m"][1]
            parameters["density_kg_m3"] = full["density_kg_m3"][1]
        elif index == 127:
            parameters["height_m"] = full["height_m"][1]
            parameters["traction_amplitude_pa"] = full["traction_amplitude_pa"][1]

        if role == "excitation_ood":
            sub_factor = 0.35 + 0.15 * unit[index, 0]
            near_factor = 1.20 + 0.20 * unit[index, 1]
            chirp_start = 0.25 + 0.10 * unit[index, 2]
            chirp_end = 2.40 + 0.20 * unit[index, 3]
        else:
            sub_factor = 0.65 + 0.20 * unit[index, 0]
            near_factor = 0.90 + 0.20 * unit[index, 1]
            chirp_start = 0.40 + 0.20 * unit[index, 2]
            chirp_end = 1.80 + 0.40 * unit[index, 3]
        rows.append(
            {
                "configuration_id": index,
                "case_id": f"config_{index:04d}",
                "split": split,
                "protocol_role": role,
                "parameters": parameters,
                "excitations": [
                    {"name": "half_sine_pulse", "kind": "half_sine_pulse"},
                    {"name": "sub_resonant_sine", "kind": "sine", "frequency_factor": sub_factor},
                    {"name": "near_resonant_sine", "kind": "sine", "frequency_factor": near_factor},
                    {
                        "name": "chirp",
                        "kind": "chirp",
                        "start_frequency_factor": chirp_start,
                        "end_frequency_factor": chirp_end,
                    },
                ],
            }
        )
    return rows


def case_config(global_config: dict, row: dict) -> dict:
    parameters = row["parameters"]
    length = float(parameters["length_m"])
    height = float(parameters["height_m"])
    return {
        "dataset": global_config["dataset"],
        "seed": global_config["seed"],
        "software": global_config["software"],
        "geometry": {
            **global_config["geometry"],
            "length_m": length,
            "height_m": height,
        },
        "material": {
            **global_config["fixed_material"],
            "young_pa": float(parameters["young_pa"]),
            "density_kg_m3": float(parameters["density_kg_m3"]),
        },
        "dynamics": {
            **global_config["dynamics"],
            "target_damping_ratio": float(parameters["damping_ratio"]),
            "traction_amplitude_pa": float(parameters["traction_amplitude_pa"]),
        },
        "sensors": [[fraction * length, 0.5 * height] for fraction in (0.2, 0.4, 0.6, 0.8, 1.0)],
        "excitations": deepcopy(row["excitations"]),
    }


def write_design(config: dict, rows: list[dict]) -> None:
    DATA_DIR.mkdir(parents=True, exist_ok=True)
    payload = {
        "schema": "agentfem.physics-data.structural-dynamics-v1-design",
        "schema_version": "1.0.0",
        "config": config,
        "configurations": rows,
    }
    temporary = DATA_DIR / "design.json.tmp"
    temporary.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n", encoding="utf-8")
    os.replace(temporary, DATA_DIR / "design.json")
    temporary = DATA_DIR / "design.jsonl.tmp"
    temporary.write_text(
        "".join(json.dumps(row, sort_keys=True) + "\n" for row in rows),
        encoding="utf-8",
    )
    os.replace(temporary, DATA_DIR / "design.jsonl")


def valid_case_file(path: Path) -> bool:
    if not path.is_file():
        return False
    try:
        with h5py.File(path, "r") as h5:
            return (
                h5.attrs.get("schema") == "agentfem.physics-data.structural-dynamics-v1-case"
                and h5.attrs.get("schema_version") == "1.1.0"
                and bool(h5.attrs.get("quality_passed", False))
                and len(h5["trajectories"]) == 4
            )
    except OSError:
        return False


def _write_case(
    path: Path,
    row: dict,
    config: dict,
    frequencies: np.ndarray,
    analytical: float,
    trajectories: list[dict],
    quality: dict,
) -> None:
    path.parent.mkdir(parents=True, exist_ok=True)
    temporary = path.with_suffix(".h5.tmp")
    with h5py.File(temporary, "w") as h5:
        h5.attrs["schema"] = "agentfem.physics-data.structural-dynamics-v1-case"
        h5.attrs["schema_version"] = "1.1.0"
        h5.attrs["generator_fix"] = "component-preserving-kinematic-bcs"
        h5.attrs["configuration_id"] = int(row["configuration_id"])
        h5.attrs["case_id"] = row["case_id"]
        h5.attrs["split"] = row["split"]
        h5.attrs["protocol_role"] = row["protocol_role"]
        h5.attrs["quality_passed"] = bool(quality["passed"])
        h5.attrs["config_json"] = json.dumps(config, sort_keys=True)
        h5.attrs["quality_json"] = json.dumps(quality, sort_keys=True)
        modal = h5.create_group("modal")
        modal.create_dataset("frequencies_hz", data=frequencies)
        modal.attrs["analytical_first_frequency_hz"] = analytical
        modal.attrs["first_frequency_relative_error"] = quality[
            "first_frequency_relative_error"
        ]
        root = h5.create_group("trajectories")
        common_written = False
        for trajectory in trajectories:
            group = root.create_group(trajectory["name"])
            group.attrs["spec_json"] = json.dumps(trajectory["spec"], sort_keys=True)
            for key in (
                "dt_s",
                "steps",
                "rayleigh_mass_coefficient_per_s",
                "rayleigh_stiffness_coefficient_s",
                "wall_time_s",
                "dof_count",
                "kinematic_bc_scalar_dof_count",
                "energy_balance_reference_j_per_m",
            ):
                group.attrs[key] = trajectory[key]
            for key in (
                "time_s",
                "force_scale",
                "sensor_displacement_m",
                "strain_energy_j_per_m",
                "kinetic_energy_j_per_m",
                "external_power_w_per_m",
                "damping_power_w_per_m",
                "external_work_j_per_m",
                "damping_dissipation_j_per_m",
                "energy_balance_residual_j_per_m",
                "energy_balance_relative_residual",
            ):
                group.create_dataset(key, data=trajectory[key], compression="gzip", shuffle=True)
            field = read_field_history(trajectory["field_hdf5"])
            if not common_written:
                common = h5.create_group("common")
                common.create_dataset("reference_geometry_m", data=field["reference_geometry_m"])
                common.create_dataset("topology", data=field["topology"])
                common.create_dataset("sensor_coordinates_m", data=np.asarray(config["sensors"]))
                common_written = True
            field_group = group.create_group("fields")
            for key in ("time_s", "displacement_m", "velocity_m_per_s", "acceleration_m_per_s2"):
                field_group.create_dataset(key, data=field[key], compression="gzip", shuffle=True)
    os.replace(temporary, path)


def run_case(global_config: dict, row: dict, *, force: bool = False) -> dict:
    path = CASE_DIR / f"{row['case_id']}.h5"
    if not force and valid_case_file(path):
        return {"case_id": row["case_id"], "status": "reused", "path": str(path)}
    config = case_config(global_config, row)
    started = perf_counter()
    modal_result = solve_modes(config, modes=4)
    frequencies = np.asarray(modal_result.quantity("frequencies"), dtype=float)
    analytical = analytical_first_frequency(config)
    frequency_error = float(abs(frequencies[0] - analytical) / analytical)
    with tempfile.TemporaryDirectory(prefix=f"t4_{row['case_id']}_", dir=DATA_DIR) as scratch:
        scratch_path = Path(scratch)
        trajectories = [
            run_trajectory(
                config,
                spec,
                frequencies_hz=frequencies,
                output_dir=scratch_path,
                save_fields=True,
            )
            for spec in config["excitations"]
        ]
        energy = {
            item["name"]: float(np.max(np.abs(item["energy_balance_relative_residual"])))
            for item in trajectories
        }
        quality = {
            "first_frequency_relative_error": frequency_error,
            "maximum_energy_balance_relative_residual": max(energy.values()),
            "trajectory_energy_balance_relative_residual": energy,
            "all_finite": bool(
                all(
                    np.all(np.isfinite(item["sensor_displacement_m"]))
                    and np.all(np.isfinite(item["energy_balance_relative_residual"]))
                    for item in trajectories
                )
            ),
        }
        quality["gates"] = {
            "analytical_frequency": frequency_error
            <= float(global_config["quality"]["maximum_first_frequency_analytical_relative_error"]),
            "energy_balance": quality["maximum_energy_balance_relative_residual"]
            <= float(global_config["quality"]["maximum_energy_balance_relative_residual"]),
            "finite": quality["all_finite"],
        }
        quality["passed"] = bool(all(quality["gates"].values()))
        if not quality["passed"]:
            raise RuntimeError(f"Quality gate failed for {row['case_id']}: {quality}")
        _write_case(path, row, config, frequencies, analytical, trajectories, quality)
        if int(row["configuration_id"]) in set(global_config["storage"]["native_field_representatives"]):
            destination = REPRESENTATIVE_DIR / row["case_id"]
            destination.mkdir(parents=True, exist_ok=True)
            for item in trajectories:
                shutil.copy2(item["field_hdf5"], destination / Path(item["field_hdf5"]).name)
                shutil.copy2(item["field_xdmf"], destination / Path(item["field_xdmf"]).name)
    return {
        "case_id": row["case_id"],
        "status": "generated",
        "path": str(path),
        "wall_time_s": perf_counter() - started,
        "first_frequency_hz": float(frequencies[0]),
        "first_frequency_relative_error": frequency_error,
        "maximum_energy_balance_relative_residual": quality[
            "maximum_energy_balance_relative_residual"
        ],
        "bytes": path.stat().st_size,
    }


def main() -> None:
    parser = argparse.ArgumentParser()
    parser.add_argument("--config", type=Path, default=CONFIG_PATH)
    parser.add_argument("--start", type=int, default=0)
    parser.add_argument("--stop", type=int)
    parser.add_argument("--design-only", action="store_true")
    parser.add_argument("--skip-design-write", action="store_true")
    parser.add_argument("--force", action="store_true")
    args = parser.parse_args()
    config = load_config(args.config)
    rows = configuration_design(config)
    if not args.skip_design_write:
        write_design(config, rows)
    if args.design_only:
        print(json.dumps({"status": "design_written", "configuration_count": len(rows)}, indent=2))
        return
    stop = len(rows) if args.stop is None else min(args.stop, len(rows))
    for row in rows[max(0, args.start):stop]:
        print(json.dumps(run_case(config, row, force=args.force), sort_keys=True), flush=True)


if __name__ == "__main__":
    main()