Data‐driven model for estimating seismic performance capacity of masonry‐infilled steel structures

Abstract

Steel structures with infill masonry walls (IMWs) are common in residential and commercial buildings, where the infill is often treated as non‐structural but in reality, has a significant impact on overall behaviour. IMWs contribute additional stiffness and strength, particularly under lateral loads, but they also introduce nonline‐arities, irregular response patterns, and potentially brittle damage mechanisms like shear cracking and out‐of‐plane failure, which can considerably affect the behaviour of structural members. Since computationally complex modelling and analysis may need a significant time and computational cost, capturing these effects accurately is challenging. In this context, stacked machine‐learning (ML) models offer a powerful alternative, as they can efficiently learn complex nonlinear relationships between material properties, geometry, and structural conditions; and then predict the main structural responses of maximum interstory drift ratio (IDRmax), maximum residual interstory drift ratio (RIDRmax), and their distributions across a wide range of scenarios. The proposed stacked ML models can estimate the IDRmax and RIDRmax, and use them for estimating their distributions along floor levels. As presented, Stacked ML‐5 showed the accuracy of 96.4% and 95.7% for estimating distribution of IDR and RIDR of the 2‐story steel MRF, and the accuracy of 97.8% and 94.1% for estimating distribution of IDR and RIDR of the 4‐story steel MRF, respectively, which considerably surpass the conventional formula‐based estimations.