Machine learning-based prediction of seismic response of steel diagrid systems
Vahid Jahangiri, Mohammad Reza Akbarzadeh, Sina Abdolrahimi Shahamat, Ali Asgari, Babak Naeim, Faramarz Ranjbar
Abstract
This study explores the use of supervised machine learning (ML) algorithms to predict the maximum inter-story drift ratio (M-IDR) of steel diagrid structures, significantly reducing the need for computationally intensive modeling. Unlike previous studies, this research focuses on creating a robust surrogate model tailored for diagrid systems, leveraging incremental dynamic analysis (IDA) data for varying building heights (4, 8, 12, 16, and 24 stories). Twenty-one ML algorithms were trained and evaluated on a dataset featuring critical structural parameters such as structural weight, fundamental period (T1), record sequence number (RSN), and spectral acceleration (Sa(T1)). Results highlight that Extra Trees Regressor, Bagging Regressor, Random Forest, Stacking Regressor, and K-Nearest Neighbors achieved superior predictive performance, with R² values exceeding 0.95 and low mean squared errors (MSE). The novelty of this work lies in its emphasis on optimizing ML algorithms for seismic vulnerability assessment of diagrid systems, addressing gaps in prior research. This approach provides an efficient and reliable alternative to traditional seismic response analyses, offering designers a practical tool to assess seismic risks in diagrid structures with improved accuracy.