CS8903 (Guided study and research)-Spring 22'

ML Surrogates for characterizing fracture toughness

Faculty mentored research project aiming to provide ML alternative to computationally intensive finite-element modelling for characterizing specimen’s fracture toughness.

Professors:

• Christos Athanasiou
• Elizabeth Qian

Relevant research papers:

• https://www.sciencedirect.com/science/article/abs/pii/S1359645420302032
• https://www.pnas.org/doi/abs/10.1073/pnas.2104765118

Methodology

Figure work in progress

File organization

-main.py (Main file with the core function calls and hyperparameter definition)
-data_func.py (File with functions/classes relevant to pre-processing of the data)
-epoch_func.py (File with functions/classes relevant to epoch-specific steps, example: training, testing, early-stopping, etc.)
-loss_func.py (File with functions/classes relevant to calculating different loss/accuracy criteria)
-model_func.py (File with functions/classes relevant to pre-defined neural network structures)
-post_func.py (File with functions/classes relevant to post-processing of data, example: plotting loss curves, accuracy curves, etc.)
-understand_data.py (File to plot the features and target values for better understanding the data's characteristics)
-setup.py (File with relevant info for running the *.py files)