Neural Operators for Stochastic Modeling of System Response to Natural Hazards

Somdatta Goswami, Dimitris Giovanis, Bowei Li, Seymour Spence, Michael D. Shields

Paper • Data

---

Overview

We propose and evaluate neural operator frameworks (self-adaptive FNO and DeepFNOnet)to model stochastic response of nonlinear structural systems under seismic and wind excitations. The approach is benchmarked against traditional simulators and surrogate models, achieving high accuracy and generalizability under uncertainty.

---

Data Access

You can download the dataset used in this study from the link below: Download Data

Contents:

• raw_data/: OpenSees simulation outputs (earthquake and wind).
• preprocessed_data/: Ready-to-use data for model training and testing.
• additional_results/: Results related to DeepFNO on wind datasets.
• Preprocessing scripts are provided in scripts/preprocess/.
• For results and visualizations of the DeepONet+FNO in the wind example, refer to: data/additional_results/.

---

Installation

Clone the repo and set up the environment:

git clone https://github.com/YOUR-USERNAME/neural-operators-hazard-response.git
cd neural-operators-hazard-response

---

Citation

If you use this code or data in your work, please cite:

@article{goswami2024neural,
  title={Neural Operators for Stochastic Modeling of Nonlinear Structural System Response to Natural Hazards},
  author={Goswami, Somdatta and Giovanis, Dimitris and Li, Bowei and Spence, Seymour and Shields, Michael D},
  journal={arXiv preprint arXiv:2502.11279},
  year={2024}
}

Contact: For questions please email Somdatta Goswami (somdatta[at]jhu.edu)