JanvDelden/vibrating_plates

Dataset for "Learning to Predict Structural Vibrations" paper

Overview

The dataset contains numerical simulations of plate vibrations given a harmonic excitation with different excitation frequencies. Between samples, the plate geometry is varied alongside material, boundary condition, load position and sizes.

This repository contains two variants:

• V5000: Here, only the plate geometry is varied by imposing a beading pattern, which represents an indentation into the plate surface. Material, boundary condition, size and load position are kept fixed.
• G5000: In addition to plate geometry, scalar physical parameters are varied. For each variant, 5000 training samples and 1000 test samples each containing numerical simulations for 1 - 300 Hz, are included. In addition, another dataset in the V5000 format is included. This dataset contains 50,000 samples with only 15 numerical simulations per sample in the 1 - 300 Hz range.

Data structure

The data is saved in the hdf5 file format and compressed with the "Blosc" compression filter from the hdf5plugin python package. Include "import hdf5plugin" to be able to read the files. The following files are included:

• Training data for V5000: V5000_2000_train.h5, V5000_3000_train.h5. Both together form the whole training dataset.
• Training data for G5000: G5000_BCFP_2000_train.h5, G5000_BCFP_3000_train.h5. Both together form the whole training dataset.
• Test data for the respective variants: G5000_BCFP_1000_test.h5, V5000_1000_test.h5
• 50,000 samples file: V50000_15f.h5

The following keys are available in each file:

• bead_patterns: The variation of the plate geometry encoded as a height map (unit: meters).
• z_vel_mean_sq: The frequency response (unit: decibels).
• z_vel_abs: The magnitude of velocity fields perpendicular to the plate (unit: meters per second).
• frequencies: The frequencies for frequency response and velocity fields (unit: hertz).
• phy_para: An array of scalar physical properties in following order:
  • plate length in range [0.6, 0.9] meters.
  • plate width in range [0.4, 0.6] meters.
  • plate thickness in range [0.002, 0.004] meters.
  • damping loss factor in range [0.01, 0.03].
  • rotational stiffness at boundary in range [0.0, 100] Nm/rad.
  • excitation point x-coordinate in range [0.2, 0.8] (unitless/normalized)
  • excitation point y-coordinate in range [0.2, 0.8] (unitless/normalized).

phy_para is varied only in the G5000 dataset. It does not need to be used for the V5000 dataset files.

Additional Information

The dataset is described in detail in this publication:

https://proceedings.neurips.cc/paper_files/paper/2024/hash/bbd7d8bd780fcf7143add2317ba04638-Abstract-Conference.html

Code is available here:

https://github.com/ecker-lab/Learning_Vibrating_Plates/tree/main