Unreinforced masonry (URM) structures are widespread worldwide, particularly in older urban districts. However, URM buildings—particularly those constructed before the introduction of modern building codes—are highly vulnerable to seismic hazard, and prone to experiencing local and/or global failures when subjected to significant horizontal shaking. Debris from seismic collapses poses a serious risk to occupants and passersby, while significantly compromising the functionality of road networks and critical infrastructure after major earthquake events. Accurate prediction of debris distribution can inform post‐earthquake risk assessment and resilience planning, enabling prioritization of emergency routes and more effective recovery strategies. Nonetheless, quantifying the extent and distribution of URM debris through laboratory testing is impractical, costly, and may cause damage to delicate instrumentation. In this paper, experimentally validated discontinuum models in Bullet developed within a computationally efficient physics‐engine‐based framework are firstly used to simulate until collapse the seismic behavior of idealized URM assemblies with varying complexities. Then, after assembling a large numerical dataset of debris predictions using combinations of ground motions and building configurations (one‐ and two‐storey cases), machine learning models are trained to generate virtual debris distributions in the form of heatmaps. A total of nine machine learning algorithms are evaluated, where gradient‐boosted decision trees (GBDT) and XGBoost achieve the best performance and are employed to analyze the relative importance of input parameters and perform parametric studies. Results have revealed patterns in how seismic debris predictions are related to both building geometry and the nature of the earthquake signals applied. The developed framework offers a versatile and computationally efficient tool for quantifying debris impact in resilience assessment and enhancing post‐earthquake recovery decision‐making strategies. It can facilitate converting debris heatmaps into roadway‐blockage probabilities for specified clearance criteria, enabling rapid clearance prioritization, safe‐route planning, and network‐level road functionality assessment.
Liu et al. (Sun,) studied this question.