Unsupervised Machine Learning-Driven Assessment of Infill Wall Systems for Structural Performance in Soft Stories

Abstract

Infill wall systems have been extensively studied as potential solutions to mitigate the adverse effects of stiffness irregularity in soft-storey structures. This research leverages unsupervised machine learning techniques, specifically clustering algorithms, to analyze and compare the mechanical behavior of different columns in a nine-storey unsymmetrical reinforced concrete building subjected to seismic loading. The primary objective is to assess the effectiveness of various infill wall systems, including 5-inch and 10-inch wide brick walls and 5-inch, 7.5-inch, and 10-inch wide concrete walls, in enhancing structural resilience. The study employs a comprehensive data-driven approach, incorporating ETABS modeling, data preprocessing, exploratory data analysis, and clustering to identify patterns and relationships in the structural performance of columns. Key findings indicate that more comprehensive and concrete infill walls significantly reduce displacement values, improving structural stiffness. Cluster analysis reveals that columns connected to multiple infill walls, particularly exterior corner columns, exhibit enhanced structural performance. Specifically, the 10-inch comprehensive concrete infill wall system demonstrated the highest efficacy in mitigating stiffness irregularity. The research further highlights that clustering algorithms, such as K-Means, effectively categorize columns based on their mechanical responses, facilitating a comparative evaluation of infill wall systems. These insights provide valuable recommendations for designing and retrofitting soft-storey structures to enhance seismic resilience.