Corrosion prediction models for chemical systems based on data and mechanism fusion: a review

Abstract

Corrosion is one of the most critical issues affecting the safe and stable operation of chemical systems. Therefore, accurate prediction of corrosion rates is essential for the safety and stability of chemical systems. Traditional mechanism-based models are challenging to construct as they struggle to account for all influencing factors and often lack sufficient accuracy when applied in chemical plants. While data-driven approaches have been widely adopted in the field of corrosion prediction, they suffer from inherent limitations such as poor generalisability, lack of interpretability and a high dependency on data quality. These limitations underscore the urgent need for developing precise and interpretable corrosion rate prediction models that integrate mechanistic insights, thereby enabling effective maintenance of equipment condition. This article systematically reviews the evolution of corrosion prediction models, from those based on kinetic, thermodynamic and electrochemical mechanisms to data-driven models leveraging multi-source data from chemical systems, and finally to mechanism-data fusion models that offer high accuracy and broad generalisability. The article introduces five fusion mechanisms – embedded, inductive, weighted, cascaded and augmented – that integrate the strengths of both mechanistic and data-driven models to enhance predictive accuracy and interpretability. These fusion mechanisms effectively combine the advantages of mechanistic models and data-driven approaches, leading to improved corrosion rate prediction accuracy and enhanced model generalisation. The article also provides a summary of existing models in the field of corrosion prediction and offers a forward-looking perspective on future research directions for mechanism-data fusion-driven corrosion prediction models.