SCEC Project Details
SCEC Award Number | 23186 | View PDF | |||||
Proposal Category | Individual Proposal (Integration and Theory) | ||||||
Proposal Title | A Data-Driven Site Response Module for the Broadband Platform: Development and Prototype Implementation | ||||||
Investigator(s) |
|
||||||
Other Participants |
Grigorios Lavrentiadis |
||||||
SCEC Priorities | 4d, 4b, 4c | SCEC Groups | GM, EEII, CS | ||||
Report Due Date | 03/15/2024 | Date Report Submitted | 07/31/2025 |
Project Abstract |
Data-driven methods offer a novel approach of learning the governing laws from sufficiently rich training data, while avoiding simplified assumptions that limit the realism of models developed with traditional statistical tools. In this study, we demonstrate this new paradigm by developing a Fourier Neural Operator (FNO) which modifies the outcrop ground motions from the SCEC BBP to account for the full nonlinear response of the near-surface soil layers. FNO was trained on non-linear one-dimensional wave propagation through smooth Bay Area velocity profiles using the site-response software, PySeismoSoil. A key advantage of the neural operator architecture in FNO compared to traditional neural networks, is its ability to learn the mapping between continuous function spaces as opposed to finite-dimensional sets, rendering the training and application of the model resolutions invariant (i.e., training can include input signals of different sampling frequencies without loss of information or generation of artifacts, while prediction can be performed on sampling frequencies independent of training). Verification analyses through residual and goodness of fit evaluations demonstrate that FNO can correctly estimate the nonlinear amplification for ground motions and profiles not included in the training dataset in the 0.1 to 30Ηz frequency range. By appropriately conditioning data-driven algorithms, our work demonstrates the potential of using these methods to learn increasingly complex physics and their uncertainty over the entire frequency range of engineering interest, and to modulate on demand time-histories appropriate for engineering design with high degree of realism. |
Intellectual Merit | The intellectual merit lies in the development of a machine learning nonlinear site response model for the BBP that is built using Neural operators, specifically a Fourier Neural Operator (FNO). The FNO was trained on non-linear one dimensional wave propagation through smooth Bay Area velocity profiles using the siteresponse software, PySeismoSoil. A key advantage of the neural operator architecture in FNO compared to traditional neural networks, is its ability to learn the mapping between continuous function spaces as opposed to finite-dimensional sets, rendering the training and application of the model resolutions invariant (i.e., training can include input signals of different sampling frequencies without loss of information or generation of artifacts, while prediction can be performed on sampling frequencies independent of training). The module can be applied to ground motions with higher frequency resolution, and higher strain intensity than the training dataset and still perform well so long as no additional laws of physics (failure, cracking etc) are introduced. |
Broader Impacts | By appropriately conditioning data-driven algorithms, our work demonstrates the potential of using these methods to learn increasingly complex physics and their uncertainty over the entire frequency range of engineering interest, and to modulate on demand time-histories appropriate for engineering design with high degree of realism. |
Exemplary Figure | Figure 7: GOF scores of the performance of the module comparing it to the surface to rock outcrop nonlinear site response amplification using SeismoSoil. Credits: Asimaki, Xia, Shi |
Linked Publications
Add missing publication or edit citation shown. Enter the SCEC project ID to link publication. |