Polarization Resistance Assessment of Reinforced Concrete Using a Transmission Line Based Model and Genetic Programming (RIP2026-00106)

Christopher Alexander, David Montes de Oca Zapiain, Ryan Katona

Sandia National laboratories, Honeywell International Inc

Polarization resistance measurements of steel in concrete are widely used in laboratory investigations and, to a lesser extent, in field assessments of corrosion.1However, the complex geometry of reinforced concrete complicates the interpretation of these measurements, particularly when relating the apparent response to the true interfacial polarization resistance. Commercial devices address this by incorporating current confinement methods that constrict current flow to a finite polarization area of the embedded steel, but confinement must be tailored to system geometry. For well-defined geometries such as prismatic beams, a mathematical expression2 has been developed to relate the measured polarization resistance to the true value by incorporating geometric factors and concrete resistivity. The model uses a transmission line formulation to account for current spreading beyond the small counter electrode relative to the beam length, as well as a parallel contribution from the portion of steel directly beneath the electrode. The associated errors decrease with cover thickness and are often below 10%, but they can become substantial when cover is large or when the steel’s polarization resistance is low relative to the concrete resistivity. Moreover, such analytical models cannot be easily extended to the more complex geometries common in field structures.
In this work, we explore machine learning as an alternative approach for recovering the true interfacial polarization resistance of steel in concrete from the apparent measured value, given the specimen geometry and concrete resistivity. We present a feasibility study based on beam geometries, with the aim of enabling future application to more complex structural configurations and reinforcement layouts. Training data for the machine learning model are generated using finite element simulations across a range of beam and counter-electrode dimensions. A comparison of the error distributions of the machine-learning model and the transmission-line-based model and the advantages and limitations of each are discussed.
Machine-learning methods may be useful for interpreting polarization resistance measurements in reinforced concrete structures with complex geometries typical of field conditions under the initial assumption of uniform corrosion and without the need for current confinement methods. Further work is required to also address the complexities of nonuniform corrosion.
1. Alexander, Christopher L. "Electrochemical Impedance Spectroscopy for Corrosion Diagnosis of Reinforced Concrete." Current Opinion in Electrochemistry (2025): 101790.
2. Alexander, Christopher L., Stanley C. Agbakansi, and Wesley Vitor Dantas De Carvalho Bezerra. "Addressing geometric influences on electrochemical impedance spectroscopy for accurate corrosion assessment in steel-reinforced concrete beams." Corrosion Science 247 (2025): 112703.