MAST-Corrosion: A Galvanic Corrosion Prediction System Based on Physics-Informed Graph Neural Networks

MAST-Corrosion is a physics-informed graph neural network-based galvanic corrosion prediction system that integrates physical constraints and deep learning. It models electrochemical interactions between materials via graph neural networks and is equipped with an interactive visualization interface. The project aims to address the limitations of traditional galvanic corrosion prediction methods and provide an efficient, interpretable prediction tool for the engineering materials field.

Project Source: GitHub, Original Author/Maintainer Th3Samaritan, Release Date June 7, 2026, Link https://github.com/Th3Samaritan/MAST-Corrosion

Galvanic corrosion is a common and destructive form of corrosion in the engineering materials field, existing in key infrastructure such as ships, bridges, pipelines, and aircraft. Traditional prediction methods rely on empirical formulas and finite element simulations, but face challenges like a large number of material combinations, high computational costs, and difficulty in modeling uncertainty factors in real environments.

Pure data-driven neural networks lack physical interpretability. Physics-Informed Neural Networks (PINNs) encode domain knowledge as constraints, balancing data fitting ability and physical laws, and have become an important direction to solve the above problems.

MAST-Corrosion uses graph neural networks (GNN) as its core architecture, modeling metal components as graph nodes and electrochemical connections as graph edges, which aligns with the physical nature of the problem. Built on the PyTorch Geometric framework, it has the ability to efficiently process large-scale graph structure data.

The core innovation is integrating physical constraints into the training process, including charge conservation laws, electrode kinetics equations, and Ohm's law constraints, to achieve multi-objective optimization and ensure predictions comply with physical laws. It also integrates an interactive visualization interface built with Streamlit, supporting functions such as material parameter input, geometric configuration definition, and real-time viewing of corrosion distribution.

The application scenarios of MAST-Corrosion include:

1. Engineering design phase: Quickly assess the corrosion risk of different material combinations, replace traditional trial-and-error methods, and shorten the design cycle;
2. Maintenance of existing facilities: Combine on-site monitoring data to predict the remaining life of key parts, optimize maintenance plans, and avoid resource waste or safety hazards;
3. Material selection decision-making: Support the expansion of material databases, evaluate the applicability of new materials in specific environments, and assist in the research and development of new alloys.

Technical Highlights:

1. Physical interpretability: PINNs output predictions consistent with physical laws, suitable for safety-critical applications;
2. Graph structure modeling: GNN naturally represents material interaction relationships, avoiding information loss;
3. End-to-end toolchain: The complete workflow reduces technical barriers, making it easy for non-technical users to use.

Limitations and Improvement Directions:

1. Data dependency: Requires accurate material parameters, and data for rare materials is difficult to obtain;
2. Complex environment modeling: Currently focuses on steady-state conditions, and the ability to handle dynamic environments (tides, temperature cycles) needs verification;
3. Multi-scale coupling: The cross-scale coupling problem between micro electrochemical reactions and macro structural corrosion has not yet been solved.

MAST-Corrosion is a beneficial attempt in the intersection of artificial intelligence and materials science, demonstrating how to encode domain knowledge as neural network constraints to build accurate and interpretable prediction models. With the development of industrial IoT and digital twin technologies, such physics-informed intelligent systems will play an important role in infrastructure health management.

This project provides a reference implementation for researchers in corrosion protection, materials engineering, or AI for Science, and is worthy of in-depth study and secondary development.