Prediction of aerodynamic coefficients for twisting shapes of buildings and structures based on machine learning and CFD-modelling
Introduction. Research was carried out on the application of machine learning to predict aerodynamic coefficients on buildings and structures with twisted form configurations. Data from aerodynamic simulations using numerical modelling in ANSYS CFX was used for training. The quality of predictions m...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Moscow State University of Civil Engineering (MGSU)
2024-05-01
|
| Series: | Vestnik MGSU |
| Subjects: | |
| Online Access: | https://www.vestnikmgsu.ru/jour/article/view/265 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Introduction. Research was carried out on the application of machine learning to predict aerodynamic coefficients on buildings and structures with twisted form configurations. Data from aerodynamic simulations using numerical modelling in ANSYS CFX was used for training. The quality of predictions made by various machine learning models was evaluated in comparison to numerical simulations. Conclusions related to the use of machine learning models for determining wind loads on buildings and structures are drawn.Materials and methods. Python programming language and the following libraries, Pandas, NumPy, Scikit-learn, and Matplotlib were used to analyze the obtained results and to develop the machine learning model. The study considered four machine learning methods: linear regression, decision tree, k-nearest neighbours, and random forest. Aerodynamic simulations based on numerical modelling methods in ANSYS CFX were used to generate the training data. The accuracy of different machine learning models in predicting aerodynamic coefficients was evaluated using the statistical measure of R-squared.Results. As a result of the research, a database of 217 numerical solutions was compiled for various angles of twist of the building’s form. These results include the distribution of aerodynamic pressure coefficients over the building’s surface, as well as aerodynamic force and moment coefficients (Cx, Cy, CMz) as a function of height. The data was used to train four machine learning models. The best-performing machine learning model (random forest) was verified by comparing it to the results of numerical modelling.Conclusions. Various machine learning models for predicting aerodynamic coefficients on buildings and structures were investigated. Conclusions were drawn regarding the applicability of machine learning methods as an alternative approach to determining wind loads. |
|---|---|
| ISSN: | 1997-0935 2304-6600 |