A probabilistic computational framework for the prediction of corrosion-induced cracking in large structures

Area of Science:

• Materials Science
• Computational Mechanics
• Corrosion Engineering

Background:

• Corrosion significantly reduces structural integrity by initiating cracks.
• Quantitative corrosion assessment and modeling of crack initiation are vital for reliability analysis.
• Existing models often lack the integration of complex electro-chemo-mechanical interactions and predictive capabilities.

Purpose of the Study:

• To develop a probabilistic computational analysis framework for predicting corrosion-induced crack initiation.
• To integrate phase-field modeling, machine learning, and uncertainty quantification for a comprehensive approach.
• To enable real-time prediction of corrosion morphology and crack initiation behavior.

Main Methods:

• An electro-chemo-mechanical phase-field model was adapted to simulate pitting corrosion, coupling stress with electrode chemical potential.
• A morphology-based crack initiation criterion was developed to quantify pit-to-cracking transitions.
• A spatiotemporal surrogate modeling approach using Convolutional Neural Networks (CNN) and Gaussian Process regression with a NARX architecture was implemented.

Main Results:

• The framework enables real-time prediction of corrosion morphology evolution.
• It accurately predicts crack initiation timing, location, and likelihood.
• Uncertainty quantification is integrated, providing probabilistic insights into structural reliability.

Conclusions:

• The proposed framework offers a robust method for assessing corrosion-induced structural damage.
• It facilitates model-based reliability analysis under various stress and corrosion conditions.
• This approach enhances the prediction of material degradation and structural failure.