This repository contains code and sample data used to generte the results in the manuscript of ``Predicting Mechanically Driven Full Field Quantities of Interest with Deep Learning-Based Metamodels''.
Mechanical MNIST datasets collection: http://hdl.handle.net/2144/39371
Mechanical MNIST Crack Path Github: https://github.com/saeedmhz/Mechanical-MNIST-Crack-Path
Pytorch: https://pytorch.org/get-started/locally/
Manuscript: https://arxiv.org/abs/2108.03995
Trained Models: https://drive.google.com/drive/folders/1CJngW03f6tHqDoukvGUWNcaWF4QGKTn6?usp=sharing
Train a dicplacement predictor model:
python ./displacement-prediction/train.py --lr="learning rate" --d="learning rate decay' --s="reduce learning rate after s epochs" --e="total epochs" --b="batch size" --c="first lavel channels" --m="model name" --n="network_name" --ch="continue training from a previous checkpoint"
Train a strain predictor model:
python ./strain-prediction/train.py --lr="learning rate" --d="learning rate decay' --s="reduce learning rate after s epochs" --e="total epochs" --b="batch size" --c="first lavel channels" --m="model name" --ch="continue training from a previous checkpoint"
Train a crack path predictor model:
7z x ./crack-path-prediction/sample-dataset/damage.7z
7z x ./crack-path-prediction/sample-dataset/material.7z
python ./crack-path-prediction/encoder.py
python ./crack-path-prediction/train.py --lr="learning rate" --d="learning rate decay' --s="reduce learning rate after s epochs" --e="total epochs" --b="batch size" --c="first lavel channels" --m="model name" --ch="continue training from a previous checkpoint"
Predicting crack paths with a trained model:
python ./crack-path-prediction/prediction.py --c="first lavel channels" --m="model name"
python ./crack-path-prediction/decoder.py