Application of two different DeepONet models in multi-physics solidification and additive manufacturing applications. Two DeepONet models were employed:
- A sequential DeepONet model with a GRU network. Used to capture time-dependent input functions in the solidification problem.
- A ResUNet-based DeepONet model with a ResUNet in the trunk network. Used to capture the changing designs in the LDED prediction problem.
The DeepONet implementation and training is based on DeepXDE: @article{lu2021deepxde, title={DeepXDE: A deep learning library for solving differential equations}, author={Lu, Lu and Meng, Xuhui and Mao, Zhiping and Karniadakis, George Em}, journal={SIAM review}, volume={63}, number={1}, pages={208--228}, year={2021}, publisher={SIAM} }
The implementation of the ResUNet is adapted from Jan Palase: https://github.com/JanPalasek/resunet-tensorflow
If you find our model helpful in your specific applications and researches, please cite this article as: @article{kushwaha2024advanced, title={Advanced deep operator networks to predict multiphysics solution fields in materials processing and additive manufacturing}, author={Kushwaha, Shashank and Park, Jaewan and Koric, Seid and He, Junyan and Jasiuk, Iwona and Abueidda, Diab}, journal={Additive Manufacturing}, pages={104266}, year={2024}, publisher={Elsevier} }
The training data is large in size and can be downloaded through the following UIUC Box link: https://uofi.box.com/s/m91ux8n3aiygpwu7dliaebw3p866dt0p All three models described in the paper are provided.