Abstract: This study proposes a physics-informed machine learning framework to predict the erosion-corrosion behavior of copper-nickel alloys in seawater environments. A dataset comprising 857 samples with 13 features was evaluated using nine different machine learning algorithms. Among these, AdaBoost and CatBoost demonstrated the highest predictive accuracy for the non-immersion and immersion conditions, respectively. SHAP value analysis identified alloy composition, flow velocity, and immersion history as the most influential factors. Based on the optimized model, an extended dataset consisting of 70,000 samples was generated and analyzed. The effects of various environmental factors on the corrosion behavior of 90/10 and 70/30 copper-nickel alloys were investigated. This analysis revealed the key influencing factors governing the corrosion behavior of copper-nickel alloys, and the critical flow velocities of 90/10 and 70/30 copper-nickel alloys were thus derived. This research enhances predictive reliability and provides mechanistic insights, offering valuable guidance for alloy design, seawater system operation, and maintenance strategies.

Key words: Copper-nickel alloys, Machine learning, Corrosion prediction, Erosion-corrosion