This repository contains the code for performing physics-based data augmentation and model training for the prediction of experimental autogenous shrinkage.
The basic requirement for using the files is a Python 3.6.3 environment with the packages listed in requirements.txt. It is advisable to create a virtual environment with the correct dependencies.
The work-related experiments were performed on Linux Fedora 7.9 Maipo. The code should be able to work on other Operating Systems as well, but it has not been tested elsewhere.
Here is a brief description of the files and folder content:
-
network: code for training the B5Net model from scratch. -
dataset: data used for training B5Net model. -
data augmentation.ipynb: jupyter notebook to convert the raw composition-based input into physics-based composition-based input and perform augmentation.
The code to run the B5Net model is provided in the network folder. In order to run the model, you can pass a sample config file to the dl_regressors_tf2.py from inside of your network directory:
python dl_regressors_tf2.py --config_file sample/sample-run_example_tf2.config
The config file defines all the related hyperparameters associated with the model training and model testing, such as loss_type, training_data_path, val_data_path, test_data_path, label, input_type, etc. To add customized input_type, please make changes to the data_utils.py.
The code was developed by Vishu Gupta from the CUCIS group at the Electrical and Computer Engineering Department at Northwestern University.
- Vishu Gupta, Yuhui Lyu, Derick Suarez, Yuwei Mao, Wei-Keng Liao, Alok Choudhary, Wing Kam Liu, Gianluca Cusatis, and Ankit Agrawal. "Physics-based Data-Augmented Deep Learning for Enhanced Autogenous Shrinkage Prediction on Experimental Dataset." In Proceedings of the 2023 Fifteenth International Conference on Contemporary Computing, pp. 188-197. 2023. [DOI] [PDF]
@inproceedings{gupta2023physics,
title={Physics-based Data-Augmented Deep Learning for Enhanced Autogenous Shrinkage Prediction on Experimental Dataset},
author={Gupta, Vishu and Lyu, Yuhui and Suarez, Derick and Mao, Yuwei and Liao, Wei-Keng and Choudhary, Alok and Liu, Wing Kam and Cusatis, Gianluca and Agrawal, Ankit},
booktitle={Proceedings of the 2023 Fifteenth International Conference on Contemporary Computing},
pages={188--197},
year={2023}
}The open-source implementation of ElemNet here provided significant initial inspiration for the structure of this code-base. The authors would like to thank Dr. Zdenek P Bazant and Dr. Ahmet Abdullah Dönmez for helpful discussions.
The research code shared in this repository is shared without any support or guarantee of its quality. However, please do raise an issue if you find anything wrong, and I will try my best to address it.
email: [email protected]
Copyright (C) 2023, Northwestern University.
See COPYRIGHT notice in the top-level directory.
This work is supported in part by the following grants: National Institute of Standards and Technology (NIST) award 70NANB19H005; Predictive Science and Engineering Design Cluster (PS&ED, Northwestern University); Department of Energy (DOE) awards DE-SC0019358, DE-SC0021399; NSF award CMMI-2053929, and Northwestern Center for Nanocombinatorics.