Machine learning guided minimum X-bracing design for progressive collapse prevention in RC buildings

Progressive collapse of reinforced concrete (RC) buildings after sudden column loss remains a low-probability, but high-consequence hazard, and efficient retrofit design still relies on repeated nonlinear simulation and trial-and-error. This study introduces a machine-learning-guided framework that moves progressive-collapse assessment toward direct retrofit decision support by estimating the minimum X-bracing quantity and layout required to prevent collapse. A numerical database covering RC buildings with 5, 9, 12, and 15 stories; 4, 5, and 6 m spans; three column-removal scenarios, and three removal levels was generated using nonlinear dynamic analyses in Extreme Loading for Structures based on the Applied Element Method. The minimum collapse-preventing configuration for each case was used to train six algorithm families, spanning linear, kernel, neural-network, and tree-ensemble approaches, under a nested cross-validation protocol for safe/failure classification, brace-count regression, and layout prediction. Random forest achieved the best classification performance (F1 = 0.962, ROC-AUC = 0.987), while XGBoost achieved the lowest brace-count error (MAE = 0.965 braces, R² = 0.847). A cross-validated SHAP analysis identified span length as the dominant parameter, accounting for 46.8% of the total attributed effect. Because under-prediction is a safety failure, a calibrated safety margin reduced the under-prediction rate from 25.0% to 4.6%. Layout prediction reproduced the complete bracing arrangement exactly in up to 71.6% of configurations. In a hold-out comparison with ELS, 17 of 20 configurations were reproduced exactly, whereas the three under-predicted schemes collapsed, so high-demand predictions require ELS verification. The framework offers a rapid pathway for screening and preliminary retrofit design, with nonlinear dynamic analysis as final verification.

Authors

Institutions

Publication Details

Journal
Engineering Structures
Published
2026-11-15
DOI
https://doi.org/10.1016/j.engstruct.2026.124089
Primary Topic
Structural Response to Dynamic Loads
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Machine learning guided minimum X-bracing design for progressive collapse prevention in RC buildings

Adem Doğangün, Moncef L. Nehdi, Muneeb Jadallah, Muhammed Şahin
Engineering Structures
Structural Response to Dynamic Loads
article

Machine learning guided minimum X-bracing design for progressive collapse prevention in RC buildings

Adem Doğangün, Moncef L. Nehdi, Muneeb Jadallah, Muhammed Şahin
article en

Abstract

Progressive collapse of reinforced concrete (RC) buildings after sudden column loss remains a low-probability, but high-consequence hazard, and efficient retrofit design still relies on repeated nonlinear simulation and trial-and-error. This study introduces a machine-learning-guided framework that moves progressive-collapse assessment toward direct retrofit decision support by estimating the minimum X-bracing quantity and layout required to prevent collapse. A numerical database covering RC buildings with 5, 9, 12, and 15 stories; 4, 5, and 6 m spans; three column-removal scenarios, and three removal levels was generated using nonlinear dynamic analyses in Extreme Loading for Structures based on the Applied Element Method. The minimum collapse-preventing configuration for each case was used to train six algorithm families, spanning linear, kernel, neural-network, and tree-ensemble approaches, under a nested cross-validation protocol for safe/failure classification, brace-count regression, and layout prediction. Random forest achieved the best classification performance (F1 = 0.962, ROC-AUC = 0.987), while XGBoost achieved the lowest brace-count error (MAE = 0.965 braces, R² = 0.847). A cross-validated SHAP analysis identified span length as the dominant parameter, accounting for 46.8% of the total attributed effect. Because under-prediction is a safety failure, a calibrated safety margin reduced the under-prediction rate from 25.0% to 4.6%. Layout prediction reproduced the complete bracing arrangement exactly in up to 71.6% of configurations. In a hold-out comparison with ELS, 17 of 20 configurations were reproduced exactly, whereas the three under-predicted schemes collapsed, so high-demand predictions require ELS verification. The framework offers a rapid pathway for screening and preliminary retrofit design, with nonlinear dynamic analysis as final verification.

Engineering StructuresVol. 371
Bursa Uludağ Üni̇versi̇tesi̇ (TR), University of Guelph (CA)
Structural Response to Dynamic Loads
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon