Stability Assessment of Concrete Gravity Dams via Multifidelity Surrogate Models
When many repetitions of an expensive or time-consuming analysis are needed, simplified models are usually adopted to reduce the cost. This is often the case with gravity dams under seismic load, especially if geometry variation needs to be considered. Deterministic analysis of dams is an important...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2025-01-01
|
| Series: | Advances in Civil Engineering |
| Online Access: | http://dx.doi.org/10.1155/adce/1229062 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | When many repetitions of an expensive or time-consuming analysis are needed, simplified models are usually adopted to reduce the cost. This is often the case with gravity dams under seismic load, especially if geometry variation needs to be considered. Deterministic analysis of dams is an important part of preliminary analyses but generally leads to overconservative designs. In recent years, many researchers have studied the potential of machine learning techniques to reduce the computational burden of dam assessment. However, generating the training dataset for a surrogate model based on high-fidelity (HF) data can be expensive when a large set of uncertain parameters is considered. To address this issue, this study proposes the use of multifidelity surrogate (MFS) models. In this method, datasets with different levels of fidelity are combined to generate a highly accurate surrogate model at a lower cost. To illustrate this, the seismic behavior of a gravity dam is assessed by means of a HF nonlinear finite element model that considers geometric, material, and seismic uncertainties. In addition, five lower fidelity (LF) models are combined with HF samples to generate multifidelity models. The goodness of fit of the models and the computational time to produce the dataset are used to identify the combination that optimizes the MFS model performance. The results show that including medium- or low-fidelity samples improves the predictive performance of a surrogate model and reduces its computational burden. The results also show that the data generation and the selection of the best LF model depend on the size of the HF dataset. |
|---|---|
| ISSN: | 1687-8094 |