Supplementary Data For “Efficient Two-Stage Sequential Machine Learning Model for Progressive Damage Analysis in Fiber-Reinforced Composites”

These files contain the data and code that was used in the study “Efficient Two-Stage Sequential Machine Learning Model for Progressive Damage Analysis in Fiber-Reinforced Composites.” This study extended NASA’s Practical Micromechanics approach and added trained Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) Surrogate Models to be able to perform Multiscale Progressive Damage Analysis on Composite Laminates. These files include training and validation data for both CNN and LSTM based surrogate models, as well as the High Fidelity Method of Cells (HFGMC) which were needed to generate the same results that were produced from the surrogate models.

The codebase that was developed uses a combination of Multiscale Composite Analysis of Practical Micromechanics and Machine Learning Inference. The scripts included have the capability to run baseline physics-only simulations, invoke the trained surrogate models, and compare surrogate model results to high fidelity results. All data and code are being released so that the methods reported in the paper can be reused and supported for purposes of transparency and reproducibility.