.png){kind=icon}
.png){kind=icon}
Return
Knowledge-graph-based bayesian analysis framework for construction safety analysis: taking collapse accident as an example
.png){kind=icon}
PRE
AI
DOI：10.1016/j.ress.2026.112780.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Construction safety involves multi-factor uncertainty and complex causal interactions that challenge conventional risk-analysis methods. This article proposes a knowledge-graph-driven Bayesian network framework, complemented by finite element (FE) simulation, to integrate structured knowledge representation with probabilistic inference and mechanism-based validation. A dataset encompassing 1164 officially documented construction accidents was utilized to develop a multi-level knowledge graph encoding project attributes, environmental conditions, and hierarchical causal factors. This knowledge graph offers a data-driven foundation for Bayesian network (BN) structure development, thereby decreasing dependence on subjective prior specifications. BN parameters were estimated using maximum-likelihood estimation, and predictive reliability was assessed through k-fold cross-validation as well as layer-wise ablation experiments to determine the incremental contributions of various causal levels. The results demonstrate that foundation pits exhibit the highest probability of collapse (46.55%), with C11 (Safety duties not enforced) identified as the primary managerial factor (51.97%). Sensitivity analysis of the alignment between FE-based deformation responses and BN-based probabilistic inference confirms that geotechnical-condition-related factors and deviations in construction sequence significantly contribute to excavation risk. The framework allows for incremental graph expansion and Bayesian parameter updating as new accident data become available. The integrated KG-BN-FE framework provides a fullchain model from risk identification to consequence prediction, offering a scientific, data-driven foundation for proactive safety management and targeted risk control in complex construction systems.
Keywords:
Construction safety
Knowledge graph
Bayesian network
Finite element simulation
Risk analysis
Journal
R
IF:
11
Papers:
805
Citations:
0