Skip to main content Skip to secondary navigation

Multi-fidelity Simulations of Wind Loading

Main content start

by Mattia Ciarlatani

Motivation and Objective

Large-eddy simulations (LES) can play a major role in estimating wind loading on buildings. Potential advantages of LES are numerous: simulations can be performed full scale, several different designs can be evaluated in parallel, and Uncertainty Quantification (UQ) can be naturally embedded to account for multiple sources of uncertainty in the boundary conditions. However, the computational cost required to evaluate wind loading under all wind directions by means of LES can be prohibitive. Our aim is to enable the use of LES as a routine design tool for the urban wind assessment by combining high- and low-fidelity simulations to reduce the computational cost of the analysis.

Methodology and Results

We are exploring different approaches to combine a large number of cheap Low Fidelity (LF) samples with a small number of expensive High Fidelity (HF) sample to obtain accurate predictions under all wind directions.

The figure below shows an example of a result obtained using a multi-fidelity Kriging approach for the root-mean-square pressure coefficient on a high-rise building at a 10o wind direction. Five high-fidelity (fine resolution) LESs, for the 0o, 20o, 40o, 60o and 80o wind directions, were used in combination with ten low-fidelity (low resolution) LESs for 0o to 90o with a 10o resolution. Across all 'unseen' wind directions, the multi-fidelity prediction has a 30% lower root-mean-square error compared to the low-fidelity prediction.

Pressure coefficients for various QoIs, methods

In ongoing work, we are exploring how to leverage the low fidelity LESs at many wind directions to identify the optimal wind directions at which the high fidelity LESs should be run. We are also exploring alternative strategies to combine the low- and high-fidelity data.

Acknowledgements & Resources

This material is based upon work supported by the National Science Foundation CAREER award number 1749610 and used the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by National Science Foundation grant number CI-1548562.