- 56 views
- 70 downloads
Surrogate Models for Hysteresis Response Prediction of Steel Braces under Seismic Loading
-
- Author / Creator
- Pessiyan, Sepehr
-
Numerical simulation is extensively used in advanced analysis of structures under seismic loading. Even though computational power and solution algorithms have advanced over the years, response evaluation of complex structures using numerical methods can still be challenging due to high computational cost, uncertainties associated with material properties, boundary conditions and numerical elements, access to advanced commercial software packages, and lack of scalability. The development of advanced techniques, including machine learning (ML) methods, along with abundant extensive laboratory test data linked to diverse structural elements, has opened new avenues for structural analysis. These methods offer potential solutions to the problems encountered in numerical simulation.
This M.Sc. thesis aims to develop data-driven surrogate models for the prediction of nonlinear hysteresis response of braces in steel concentrically braced frames and steel buckling-restrained braced frames under seismic loading using artificial neural networks powered by the long short-term memory (LSTM) algorithm. These surrogate models are intended to be used in nonlinear seismic analysis of steel braced frame structures. The data-driven models are designed using two approaches: 1) the first approach estimates the hysteresis response parameters of the brace using LSTM architecture trained on laboratory test data and synthetic numerical data of steel braces, namely tensile yielding force capacity, compressive buckling and post-buckling capacities are estimated using the surrogate models. 2) the second approach expands the application of the proposed LSTM model to predict the complete axial load time history of steel braces using transfer learning methodology leveraging the knowledge learned by the initially trained LSTM model. The surrogate models developed using these approaches are validated using laboratory and synthetic data. Particularly, static analysis of isolated braces and pseudo-dynamic hybrid simulation of a complete steel braced frame structure subjected to earthquake accelerations are performed.
The findings suggest that the proposed surrogate models offer a computationally efficient technique with sufficient accuracy to conduct both nonlinear static and nonlinear dynamic analyses of steel braced frame structures under seismic loading. Moreover, the application of transfer learning, as an innovative approach for nonlinear hysteresis prediction in steel structures, is demonstrated to bypass the complexity associated with constructing response prediction surrogate models. -
- Graduation date
- Fall 2024
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.