LIFE-CYCLE MAINTENANCE OPTIMIZATION OF CORRODED STRUCTURE BY MACHINE LEARNING-BASED SEISMIC RELIABILITY ANALYSIS

Abstract

Life-cycle optimization in civil engineering balances cost and structural performance. It designs reliable, cost-effective strategies to counteract failures from hazards such as earthquakes. However, it is challenging to consider uncertainties from material variability, aging, and environmental factors throughout the entire life cycle. Therefore, the integration of reliability analysis is crucial for a consistent assessment under uncertainties, ensuring resilient infrastructure. This thesis examines material degradation and earthquake impacts, utilizing Kaul's model and Chinese codes for real-world ground motion simulation. This thesis also takes corrosion effects on seismic resilience into account, simulating corrosion depth using a random field for precise spatial-temporal variability modeling. Machine learning techniques, such as MLPR, GBRT, XGBoost, combined with Subset Simulation, offer an efficient method for evaluating life-cycle performance. The Genetic Algorithm determines optimal strategies with the lowest cost while ensuring structural performance. The integration of entire computational process is illustrated through a concrete frame structure example.