Research on Corrosion Prediction Model Based on Mechanism and Data Fusion

Abstract Frequent corrosion incidents of equipment and process pipelines in the refining and chemical production process pose a serious threat to the safe and stable operation of the equipment. Corrosion prediction, as the core of corrosion risk management, currently lacks a theoretical prediction model accurately derived from fundamental mechanisms based on data generation. Key challenges include unclear corrosion mechanisms, ineffective data fusion strategies, and limited precision in existing predictive models. With the development of artificial intelligence, exploring unknown and potential corrosion patterns and information under complex conditions based on the cross-fusion of mechanisms and data has become a key direction for corrosion prediction. This paper proposes an interpretable, widely applicable, and high-precision mechanism-data fusion corrosion prediction method. A corrosion kinetics model for typical coupled corrosion environments is developed through mechanistic analysis, and the application corrosion dataset is augmented under the guidance of the mechanism model using K-Nearest Neighbors (KNN) and Generative Adversarial Network (GAN) techniques. Subsequently, an optimized random forest-based corrosion rate prediction model is established. Through the application verification of on-site data of process pipelines in a low-temperature coupled corrosion environment, the corrosion rate prediction model constructed in this paper has good predictive performance, with an RMSE of 0.00623, an MAE of 0.00452, and an R2 of 0.816. This method has certain theoretical significance for improving the precision and applicability of corrosion prediction, and can provide technical support for equipment risk management and predictive maintenance, ensuring the safe and reliable operation of the equipment.