UC San Diego

Corrosion, particularly pitting corrosion, stands as one of the most prevalent forms of deterioration in large infrastructure, known for its slow, continuous, capacity-degrading nature. With the aging of structures, there is an increasing demand for the assessment and monitoring of corrosion status to facilitate system life-cycle management and optimal maintenance. However, due to the intricacies involved in the corrosion process, several challenges impede effective diagnosis and prognosis. These challenges include the inadequate coupling of mechanical stress into existing corrosion simulations, the computational intensity of most physics-based corrosion simulations limiting probabilistic studies and real-time predictions, and the difficulty in detecting pitting corrosion due to its localized nature, making it impractical to measure individual pits. This dissertation presents a comprehensive prognosis and diagnosis framework applied to a miter gate case study. The framework integrates multi-scale simulation including mesoscale Phase-Field (PF) simulation and macroscale structural analysis with Machine Learning (ML) methods to enable real-time corrosion diagnosis and prognosis. It enables for effective structural health monitoring (SHM) with digital twin (DT) for corrosion damage. The key components of this framework encompass the development of a multi-scale simulation for simulating pitting corrosion in large structures, the construction of ML-based surrogate models to expedite simulations, the implementation of uncertainty quantification (UQ), and the integration of pitting corrosion diagnosis and prognosis.