Electrochemical optimization of soil corrosion map with machine learning

Purpose This study aims to fill the research gap on carbon steel corrosion in acidic soil and optimize the corrosion map of Q235 steel in Hunan Province, China, by integrating short-term electrochemical measurements with machine learning. Design/methodology/approach Electrochemical impedance spectroscopy tests were performed on Q235 steel in saturated acidic soils from 50 sites across Hunan Province after 2 h and 120 h of immersion. Key influencing factors were identified using random forest analysis and Spearman correlation. Impedance values and annual corrosion rates were predicted and spatially mapped using ordinary least squares, random forest and support vector regression models. Findings The random forest model performed best for short-term impedance prediction (2 h), while the ordinary least squares model was more accurate for long-term prediction (120 h). A strong linear relationship (R2 = 0.913) existed between the short-term impedance predicted by random forest and the corrosion rate. HCO3– was the most influential factor for short-term impedance (contribution 16.92%), whereas SO42- dominated long-term impedance (contribution 20.04%). Originality/value This study presents a novel and practical methodology for rapidly predicting long-term corrosion rates in acidic soils by leveraging easily obtainable short-term electrochemical data combined with machine learning, offering significant utility for corrosion risk assessment and infrastructure planning.