DynaResp

Reproducible structural-dynamics simulation and synthetic dataset generation, with explicit control over which measurements and physical truth are retained.

Python 3.11-3.13 NumPy + SciPy Incremental Zarr storage GPL-3.0-or-later

pip install dynaresp

DynaResp connects a structural model, a load, and a sensor layout through one simulation contract. The result includes measured signals, clean signals, state histories, matrices, modes, forces, configuration, and random seeds.

flowchart LR
    S[Structural system] --> E[Simulation engine]
    L[Load definition] --> E
    M[Sensor layout] --> E
    E --> R[SimulationResult]
    R --> D[StructuralDataset]
    D --> Z[Zarr]
    D --> T[PyTorch]
    D --> X[xarray / pandas]

First simulation

import numpy as np

import dynaresp as dr

time = np.arange(1000) / 100.0
positions = np.arange(3)

system = dr.systems.ShearBuilding(
    masses=[1000.0, 900.0, 800.0],
    stiffnesses=[1.0e6, 8.0e5, 6.0e5],
    damping_ratio=0.02,
)

spec = dr.SimulationSpec(
    system=system,
    load=dr.loads.WhiteNoise(position=0, rms=10.0),
    sensors=(
        dr.sensors.Accelerometer(position=0),
        dr.sensors.Accelerometer(
            position=2,
            noise=dr.sensors.GaussianNoise(snr_db=30.0),
        ),
    ),
    time=time,
    positions=positions,
)
result = dr.run_simulation(spec, seed=42)

result.signal
(2 sensors, 1000 samples) result.displacement
(3 DOFs, 1000 samples) result.force
(3 DOFs, 1000 samples) result.mode_shapes
(3 modes, 3 positions)

The same workflow applies to matrix-defined MDOF systems and the simply supported finite-element beam. Loads generate a complete field on the supplied axes; the solver projects it into generalized coordinates.

Choose a path

• Run your first response

  Install the package, build a shear model, add sensors, and inspect every output array.

  Open the quickstart

• Model a structure

  Choose assembled masses, direct matrices, a shear building, or the finite-element beam.

  Read the system guides

• Generate ML-ready data

  Sample physical parameters deterministically and write large datasets incrementally to Zarr.

  Build a dataset

• Understand the architecture

  Learn the coordinate conventions, array shapes, projection boundary, and reproducibility model.

  Read the core concepts

Design commitments

• Physics stays inspectable. Simulations expose matrices, modes, state histories, forces, and clean signals; datasets retain the selected parts.
• Components stay composable. Systems, loads, sensors, and measurement effects have separate responsibilities.
• Randomness stays reproducible. Dataset samples derive their random state from the dataset seed and sample index, independent of worker count.
• Variability stays controllable. Exact cases and parameter grids support designed experiments without imposing a probability distribution.
• Storage stays scalable. Selected arrays are written incrementally to Zarr instead of accumulating the dataset in memory.

Physical scope

DynaResp currently models linear time-invariant dynamics. Beam models use two-node bending elements followed by modal reduction. Review the physical scope before using generated data as a model of a real structure.