Speaker
Description
Ageing riveted railway bridges are increasingly affected by variable-amplitude train loading, corrosion-induced section loss, rivet deterioration, and uncertainty in visual inspection, making fatigue-risk assessment and maintenance planning complex. Current bridge management practice lacks an integrated decision-support model that can simultaneously predict component-level fatigue risk, explain the governing deterioration mechanisms, and optimize maintenance actions under conflicting safety, cost, and serviceability objectives. This study proposes PIFMO, a Physics Informed Fatigue and Maintenance Optimization model for fatigue-risk-based maintenance prioritization of ageing riveted railway bridges. The framework combines structural appraisal data, FE-validated stress histories, corrosion wastage modelling, sequential-law fatigue damage, modal-parameter variation, and riveted connection condition into a unified predictive-optimization workflow. Machine learning algorithms are trained to estimate remaining fatigue life and classify fatigue-risk levels for bridge components, while SHAP and LIME explainability is used to verify whether the model predictions are governed by physically meaningful variables. NSGA-II is then applied to identify Pareto-optimal maintenance strategies that maximize remaining fatigue life while minimizing repair cost, traffic disruption, and failure risk. The scientific novelty lies in embedding fatigue, corrosion, vibration, and connection-level mechanics into an explainable ML surrogate linked directly to maintenance optimization. The case-study results indicate that the highest-priority interventions occur where low remaining fatigue life, corrosion progression, modal sensitivity, and rivet deterioration coincide. The proposed model provides a transferable framework for risk-informed maintenance planning of ageing railway steel infrastructure.
| Topic Area | Fatigue, Corrosion and Structural Integrity of Engineering Structures and Mechanical Systems |
|---|