MeC-Meso-L

Published: 2 February 2026| Version 1 | DOI: 10.17632/3yj5g8wzsy.1
Contributors:
,

Description

Purpose: train/evaluate CNNs for full-field failure-index prediction in tow-based discontinuous composites. Contents: 4000 specimens; 224×224 pixel 2D feature maps [Ex, Ey, Gxy, Vf, c2] and a 224×224 label map Failure index (Denoted FI in dataset). Generation: mesoscale finite-element simulations under uniform remote tension with explicit tow deposition. Use: supervised dense regression; standardize features per channel (training-set stats); report RMSE/SSIM (NRMSE for cross-dataset comparison).

Files

Steps to reproduce

The zip file contains 100 parquet files with each 40 specimens. Each parquet file contains the columns: specimen, FI, Ex, Ey, Gxy, Vf, c2. specimen: index between 1 and 4000 FI: failure indices under longitudinal tension Ex: transverse modulus Ey: longitudinal modulus Gxy: shear modulus Vf: Fibre volume fraction c2: mean squared cosine of tow angle relative to the loading direction The 2D specimen data can be obtained by reshaping each file after loading into (40, 224, 224, 7), i.e. 40 specimens with 7 features on a 224x224 grid.

Institutions

Categories

Composite Material, Machine Learning, Fiber-Reinforced Composite, Convolutional Neural Network, Finite Element Modeling

Funders

Licence