Assessing the Quality of Real-Time Hybrid Simulation Tests with Deep Learning Models

Abstract

Hybrid simulation (HS) is an advanced dynamic testing method that combines experimental testing and analytical modeling simultaneously to provide a better understanding of the structural systems as well as the structural elements while maintaining cost-effective solutions. A complex analytical substructure in the HS can be challenging, especially to conduct real-time HS (RTHS) tests due to the nature of numerical solution algorithms. Therefore, alternative methods, such as machine learning models are being explored to represent the analytical substructures of the RTHS tests. This study investigates the quality of the RHTS tests when a deep learning algorithm is used as a metamodel of the analytical substructure. A one-bay one-story concentrically braced frame (CBF) is selected to be used in RTHS tests where the frame is the analytical substructure, and the brace is tested experimentally. The compact HS laboratory at the University of Nevada, Reno, was used to run the RTHS experiments. Deep long short-term memory (LSTM) networks were selected to be trained as a metamodel using the Python environment to represent the dynamic behavior of the analytical substructure CBF. The pure analytical solution of the CBF under earthquake excitation is used as a training dataset of the metamodels. Several RTHS tests were performed. The quality of the test results was evaluated against the pure analytical solutions obtained from both the finite element model (FEM) and machine learning (ML) model.