Predicting Main Characteristics of Reinforced Concrete Buildings Using Machine Learning

Abstract

This paper presents a comprehensive study of five machine learning (ML) algorithms for predicting key characteristics of Reinforced Concrete (RC) structural systems. A novel dataset, ModRes, consisting of 9723 examples derived from modal and response spectrum analyses on masonry-infilled three-dimensional RC buildings, was created for ML applications. The primary objective is to develop an ML model using five distinct algorithms from the literature, capable of concurrently predicting torsional irregularity, modal participating mass ratio (MPMR), and the fundamental period in a 3D environment, while accounting for the influence of infill walls. Additionally, the study aims to determine the applicability of pushover analysis (POA) without the need for extensive numerical modeling and analysis. This approach optimizes the preliminary design process with minimal computational effort, providing valuable insights into dynamic and torsional responses during seismic events. The Categorical Boosting algorithm demonstrated outstanding performance, achieving R2 values of 0.977 for torsional irregularity, 0.997 for the fundamental period, and 0.923 for MPMR on the test dataset. It also successfully predicted POA applicability with an error rate of only 1.36%. This study highlights the practical application of ML algorithms, underscoring their effectiveness in structural engineering.

Keywords

• machine learning
• torsional irregularity
• fundamental period
• modal participating mass ratio
• pushover applicability