Assessment of corrosion probability of steel in mortars using machine learning

Corrosion assessment enables engineers to quickly discern the corrosion status of steel in concrete structures. However, existing assessment methods mainly rely on a single-factor and exhibit poor adaptability to various corrosion scenarios. Moreover, most methods are traditional deterministic approach, which ignores the uncertainties in corrosion assessments. In this work, machine learning (ML) is employed to develop a multifactor classification model for multi-level corrosion status assessment, together with corresponding corrosion probability maps. First, a comprehensive corrosion dataset was collected, including relative humidity (RH), electrical resistivity (ER), corrosion potential (CP), and corrosion rate (CR). The CR was used to subdivide different corrosion levels, and ML classification models were established for three-factor and two-factor scenarios. The optimal model was then used to create corrosion probability maps for various corrosion levels. The results indicated that the poor reliability and accuracies in current corrosion assessment methods originated from the inconsistent corrosion behaviors induced by carbonation and chloride in concrete. Moreover, when using the corrosion probability maps to assess corrosion status of steel in mortars, CP and ER should first be used to determine if the steel is in an active state, followed by RH and CP to evaluate whether it is in a severe-corrosion state.