A multi-fidelity numerical framework for predicting wind loads on buildings
Abstract/Contents
- Abstract
- According to the Revision of World Urbanization Prospects, published by the United Nations in 2018, 2.5 billion people are expected to populate big cities by $2050$. As a consequence, the number of high-rise buildings has drastically increased over the last decades, to fulfill the need of expansion of urban areas. Not only are tall buildings increasing in number, they are also becoming taller, more slender and complex in shape, making the design process very challenging. Another recent tendency in high-rise buildings' construction is the employment of glazed panels to cover the external facades. Cladding design requires accurate estimations of wind loads to guarantee the structural integrity of the panels, without resorting to overly conservative designs. This is especially important near the corners and edges of high-rise buildings' side walls, where the wind-induced pressure can result in extreme suction loads. Nowadays wind tunnel experiments are required by the building standards to estimate the wind loads on tall buildings and their components. The procedure is well established, but limited in the number of measurements locations and spatial resolution. Computational fluid-dynamics (CFD) offers an attractive alternative to the wind engineering community, given the potential to provide the complete $3$-dimensional flow field in complex geometries with a spatial resolution equal to the grid resolution. Nevertheless, significant progress is still necessary to guarantee the accuracy of the results and extensively employ CFD simulations for design purposes. In addition, the computational cost should be reasonable to practically use the simulations in the design process. The accuracy of CFD simulations of atmospheric boundary layer (ABL) flows can be compromised by two factors: uncertainties in the operating parameters, such as the boundary conditions, and simplified model assumptions, such as the turbulence model. The presence of these uncertainties is limiting the use of CFD to qualitative analysis in early design phases. This could be addressed by incorporating methods for uncertainty quantification (UQ) in the predictive models. By providing predictions in terms of mean values and confidence intervals, the models can provide significantly more meaningful information to designers and engineers than the outcome of a single deterministic simulation. Therefore, the long-term goal of this research is to enable the extensive use of CFD simulations throughout the different phases of high-rise building's design, by improving the reliability and efficiency of the numerical simulations. To achieve this goal, two CFD models with different level of fidelity and computational cost are employed: large-eddies simulations (LES) and Reynolds-averaged Navier-Stokes (RANS) simulations. Validation with wind tunnel data is carried out at all stages, to evaluate the performance of the proposed numerical framework. LES can directly provide the main quantities of interest (QoIs) of cladding design, i.e. the peak pressure distribution around the building and the design pressure of the panels. However, the generation of proper inflow conditions can be challenging and result in significant uncertainty in the results. Therefore the first goal of this thesis is to address this challenge and improve the confidence in the LES prediction of the pressure distribution around high-rise buildings. As a first step, an inflow generation method able to accurately produce user-specified turbulence statistics, while maintaining low computational cost, is developed. The method is the starting point of a LES UQ framework, that has not only the goal to produce desired ABL statistics, but also to account for the inherent uncertainty in the definition of these target profiles. The framework provides a systematic way to address the great sensitivity to the inflow conditions, that LES on high-rise buildings manifested in literature. The results demonstrate that, if the incoming boundary layer is properly generated and the underlying uncertainty is considered, LES can provide accurate prediction of the QoIs of cladding design. The computational cost of LES simulations can be prohibitive for wind engineering applications, especially during the early phases of design; for this reason, despite the turbulence model represents an important source of error, RANS simulations still represent an attractive tool. In this regard, the second objective of this thesis is to improve the confidence in the RANS prediction of the mean pressure distribution around the building, by quantifying the combined effect of uncertainties in the inflow conditions and turbulence model. The results confirm the well know RANS limitations in dealing with the flow around bluff bodies and the great level of uncertainty introduced by the turbulence model. Therefore, to overcome these issues a multi-fidelity framework is established, by leveraging a limited number of LES simulations. Validation against LES at different wind directions, shows significant improvement in the mean pressure prediction, over the standard RANS method. When designing cladding systems, we are ultimately interested in the pressure fluctuations. Therefore, the third objective of the thesis is to extend the multi-fidelity framework to enable the calculation of the root mean square pressure from RANS simulations and a limited number of LES. In this regard, a machine learning framework is proposed, to relate the LES data of rms pressure coefficient to RANS time-averaged quantities. The results demonstrate that the procedure has the potential to drastically reduce the number of LES simulations needed for design, while assuring good level of accuracy.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2020; ©2020 |
| Publication date | 2020; 2020 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Lamberti, Giacomo |
|---|---|
| Degree supervisor | Gorle, Catherine |
| Thesis advisor | Gorle, Catherine |
| Thesis advisor | Dabiri, John O. (John Oluseun) |
| Thesis advisor | Kitanidis, P. K. (Peter K.) |
| Degree committee member | Dabiri, John O. (John Oluseun) |
| Degree committee member | Kitanidis, P. K. (Peter K.) |
| Associated with | Stanford University, Civil & Environmental Engineering Department |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Giacomo Lamberti. |
|---|---|
| Note | Submitted to the Civil & Environmental Engineering Department. |
| Thesis | Thesis Ph.D. Stanford University 2020. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2020 by Giacomo Lamberti
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...