We developed a Physics-guided Convolutional Neural Network for data-driven seismic response modeling. Available physics (e.g., the laws of dynamics) are embeded into the deep learning process to guide and enhance the model training from limited measurements. With this specific design of neural network architecture considering domain knowledge, our model can further alleviate overfitting issues, reduce the need of big training datasets, and thus improve the robustness of the trained model for more reliable prediction. The performance of the proposed approach was illustrated by both numerical and experimental examples with limited datasets either from simulations or field sensing. The results show that the proposed deep PhyCNN model is an effective, reliable and computationally efficient approach for seismic structural response modeling. The trained model can further serve as a basis for developing fragility function for building safety assessment. Overall, the proposed algorithm is fundamental in nature which is scalable to other structures (e.g., bridges) under other types of hazard events.

For more information, please refer to the following:

• Zhang, R., Liu, Y., & Sun, H. (2020). Physics-guided convolutional neural network (PhyCNN) for data-driven seismic response modeling. Engineering Structures, 215, 110704.

@article{zhang2020physics,
  title={Physics-guided convolutional neural network (PhyCNN) for data-driven seismic response modeling},
  author={Zhang, Ruiyang and Liu, Yang and Sun, Hao},
  journal={Engineering Structures},
  volume={215},
  pages={110704},
  year={2020},
  publisher={Elsevier}
}