| The seismic response of Reinforced Concrete (RC) buildings is significantly influenced by the relative orientation of the earthquake motion and the building’s structural axes. Numerical damage indices (DIs), such as the Maximum Interstory Drift Ratio (MIDR), are often used to quantify the overall seismic damage level (SDL) of the building. The value of such DIs can be significantly increased if the earthquake motion is applied at the critical angle (θcr) rather than along the building’s structural axes. However, the required procedure is very computationally demanding, involving a series of Nonlinear Time History Analyses (NTHAs) for different earthquake orientations and potentially many different seismic records. In this research, novel Machine Learning (ML) methodologies were employed for the rapid estimation of the influence of θcr on the overall SDL of RC framed buildings. The problem was cast as a binary classification task and five robust models, namely Random Forest, LightGBM, XGBoost, k-Nearest Neighbors and Support Vector Machines were trained. The two best performing models, namely LightGBM and XGBoost, obtained a generalization accuracy of approximately 86%. Furthermore, they obtained an accuracy of approximately 90% in identifying cases where the computation of θcr can safely be avoided, and a corresponding accuracy of approximately 77% in identifying cases where this procedure must be performed. Thus, the proposed models provide accurate and reliable predictions, enabling engineering practitioners to significantly reduce the time and effort required to evaluate the influence of θcr on the SDL of RC framed structures and assess their overall safety. |
*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.