An interpretable machine learning-based pitting corrosion depth prediction model for steel drinking water pipelines

Steel pipes are a crucial element of the water supply system and are necessary for safely delivering large quantities of water from purification plants to consumers. Corrosion is a significant factor that deteriorates the interior and exterior of the steel pipes. Although the effectiveness of machine learning has been demonstrated in various fields, machine learning has rarely been used to identify corrosion mechanisms in buried steel pipes. A hybrid machine-learning-based corrosion depth prediction model was developed by integrating a corrosion depth trend prediction model based only on elapsed years with machine-learning algorithms. Shapley additive explanation (SHAP) was used to analyze the hybrid machine-learning-based corrosion depth prediction models, revealing corrosion mechanisms and explaining the interactions among influencing factors through global and local interpretations. The SHAP local interpretation showed that the hybrid machine-learning-based corrosion depth prediction models can effectively capture the interrelationship between soil and water corrosiveness.