Machine learning-based estimation of the out-of-plane displacement of brick infill exposed to earthquake shaking

Abstract

This study aims to develop machine learning-based prediction models for the out-of-plane displacement of infill walls under earthquake excitation by using machine learning methods and comparing them to select the best prediction. For this purpose, two shake table experiments are selected. A set of data is compiled from the selected shake table experiments that were conducted on unreinforced brick infill (URB) and bed joint reinforced infill (BJR) walls enclosed in a reinforced concrete frame (RCF). Then, machine learning models such as extreme learning machine (ELM), support vector regression (SVR), decision tree (DT), bootstrap aggregation ensemble (Bagging), and least-squares boosting ensemble (LSBoost) algorithms, were devised to estimate out-of-plane (OOP) displacements of URB and BJR. Also, the model performances were compared to each other's. The OOP displacements of URB were predicted with 0.995%, 0.983%, 0.989%, 0.994%, and 0.996% accuracy using ELM, SVR, DT, Bagging, and LSBoost, respectively. Furthermore, the ELM, SVR, DT, Bagging, and LSBoost methods predicted the OOP displacement of BJR with 0.983%, 0.568%, 0.623%, 0.606%, and 0.644% accuracy, respectively. This study found that the ELM method predicted OOP displacements of the BJR with higher accuracy than other methods. However, the LSBoost method demonstrated superior performance in estimating URB's OOP displacement.