A Real-Time Reliability Assessment Method for Urban Bridge Erecting Machines Based on a Three-State Noisy-OR Bayesian Network and XGBoost Evidence Fusion
Journal of Civil Engineering and Urban Planning · 2026 · 8 (2)
DOI: 10.23977/jceup.2026.080202 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Civil Engineering and Urban Planning, 2026.
Abstract (Originalsprache)
: The hoisting trolley system of an urban bridge erecting machine is a strongly coupled electro-mechanical-hydraulic subsystem with very few field samples for its critical components. In real-time reliability assessment, two issues are normally treated separately: low-level multi-source condition monitoring and system-level probabilistic reasoning. To bridge this gap, this paper proposes a real-time reliability assessment method based on a three-state Noisy-OR Bayesian network (BN) coupled with XGBoost evidence fusion. First, the state space of the bottom-layer Bayesian network nodes is extended from the conventional Normal/Failed binary form to a Normal/Degraded/ Failed three-state form, with a hybrid modelling strategy in which mechanical/hydraulic components are described by three states and electrical/logical components remain binary. Second, a simplified three-state Noisy-OR risk-propagation rule is adopted to aggregate the Normal, Degraded and Failed probabilities of bottom-layer nodes upward into system-level state probabilities. Finally, XGBoost is used to convert multi-source monitoring features into an initial SoftMax three-class probability vector; after soft-evidence stabilization, the corresponding prototype evidence vector is injected into the leaf nodes of the Bayesian network, closing the loop from data perception to topological reasoning. Taking a hoisting-trolley fault-diagnosis test platform as the case study, 200 days of operating data were obtained from its 3-D model and the corresponding scaled test rig. Results show that the proposed framework triggers a yellow alarm on day 74, i.e. 25 days after the first soft-fault injection and 14 days after the second. After excluding the interpretable ∼ 10-day conformation delay introduced by the soft-evidence stabilization module, the effective risk-response time is reduced to several days, and the framework escalates to a red alarm on day 78, providing a quantitative “when-to-intervene” decision basis for on-site predictive maintenance.
Zitieren
APA
(2026). A Real-Time Reliability Assessment Method for Urban Bridge Erecting Machines Based on a Three-State Noisy-OR Bayesian Network and XGBoost Evidence Fusion. Journal of Civil Engineering and Urban Planning, 8(2). https://doi.org/10.23977/jceup.2026.080202
Vancouver
A Real-Time Reliability Assessment Method for Urban Bridge Erecting Machines Based on a Three-State Noisy-OR Bayesian Network and XGBoost Evidence Fusion. Journal of Civil Engineering and Urban Planning. 2026;8(2). doi:10.23977/jceup.2026.080202
BibTeX
@article{euvimed2026,
title = {A Real-Time Reliability Assessment Method for Urban Bridge Erecting Machines Based on a Three-State Noisy-OR Bayesian Network and XGBoost Evidence Fusion},
journal = {Journal of Civil Engineering and Urban Planning},
year = {2026},
volume = {8},
number = {2},
doi = {10.23977/jceup.2026.080202}
}
EUVIMED