News Coatings Technologies
Physics-informed AI models tackle coating corrosion prediction
A critical review evaluates scientific machine learning approaches for predicting coating degradation and corrosion, identifying five persistent limitations of current data-driven models. The authors propose a new framework that combines physics-based constraints with data-driven learning to enable more reliable digital twins for asset management.
The prediction of coating degradation and corrosion is undergoing a shift from empirical regression models towards scientific machine learning (SciML), which integrates governing equations, geometry and uncertainty directly into data-driven approaches. A recent critical review examines physics-informed neural networks (PINNs), neural operators, phase-field hybrids, graph neural networks and probabilistic models with a focus on protective coatings and localised corrosion. The authors organise the literature by the mechanism through which physical knowledge enters the model, rather than by model name alone, providing a structured basis for comparison across materials, coating classes and corrosive environments.
The review identifies five persistent limitations of current SciML approaches: dimensionally inconsistent composite loss functions, incomplete enforcement of conservation and thermodynamic irreversibility, inadequate treatment of anomalous transport and material memory, loss of topological information during pit nucleation and coalescence, and weak out-of-distribution validation. These shortcomings are shown to limit the reliability of learned models when applied to real industrial data, which typically combine electrochemical impedance spectroscopy (EIS), ultrasonic thickness measurements, surface images, environmental histories and mass-loss data.
Event tip:
The EC Conference Digitalisation in Coatings Formulation is the ideal platform to learn about the latest advances and trends in the field of digitally assisted coatings production processes. Join experts, researchers and industry leaders as they share their insights and expertise in automation, big data, AI, and predictive modelling to enhance efficiency, accuracy, and speed in formulation, testing, and quality control.
Proposed framework and implications for digital twins
Building on this analysis, the authors propose a Thermo-Fractional Graph Phase-Field Neural Operator (TFG–PFNO) as a research framework rather than a validated model. The concept combines fractional memory operators, graph-based geometry representation, phase-field interface evolution, metriplectic dynamics and calibrated uncertainty quantification. Graph encoders represent material chemistry and irregular geometry, while probabilistic outputs predict both degradation fields and their uncertainty. The framework is presented as a testable hypothesis, intended to guide future implementation, ablation studies and experimental validation.
A study-level evidence map distinguishes experimentally demonstrated capabilities from those validated only against simulations or that remain conceptual. According to the authors, credible corrosion digital twins must preserve physical admissibility and propagate uncertainty to structural risk, inspection planning and maintenance decisions, rather than focus solely on interpolation accuracy. Given the estimated global cost of corrosion of approximately USD 2.5 trillion per year (around 3.4 % of global GDP), of which protective coatings absorb the largest share of mitigation effort, the review positions physics-consistent SciML as a strategically relevant direction for the coatings community.
Source: Rojas-Valdivia, L. et al., Scientific Machine Learning for Coating Degradation and Corrosion Prediction: A Critical Review and Thermodynamic Research Perspective. Coatings 16, 921 (2026).
