Hybrid data-driven machine learning approach for evaluating steel corrosion in concrete using electrical resistivity and documented concrete performance indicators

Abstract

Accurate assessment of steel corrosion in reinforced concrete is essential for ensuring durability and optimizing maintenance strategies. This study proposes a hybrid data-driven approach that integrates electrical resistivity (ER) with key concrete performance indicators—clear cover (cc), design strength (σ), and crack width (W_c)—to improve corrosion prediction. A laboratory-based dataset was generated using reinforced concrete specimens subjected to impressed current-induced corrosion. Six machine learning (ML) algorithms—Gaussian Process Regression (GPR), Support Vector Machines (SVM), Neural Networks (NN), Linear Regression (LR), Decision Trees (DT), and Bagged Trees (BT)—were employed to develop predictive models using various combinations of the four input parameters. Results show that models incorporating combined material indicators significantly outperform those using ER alone, with GPR achieving the highest predictive accuracy. The findings emphasize the value of integrating documented concrete properties to enhance the interpretation of ER measurements and support the development of practical, data-driven tools for corrosion assessment in structural health monitoring.