Physics-informed neural network inference of a strong-field-suppressed mobility law for multi-ion electromigration in concrete

Electrically accelerated migration tests and electrochemical treatments of concrete are widely used for durability assessment and repair. However, under strong electric fields, conventional constant-mobility models may introduce systematic prediction bias because they do not explicitly represent how local electric-field variation alters ionic transport behaviour. To address this issue, this study infers an electric-field-dependent ionic-mobility law for strong-field multi-ion transport in concrete and embeds it into the Poisson-Nernst-Planck framework. A physics-informed neural network with importance-adaptive training is used to infer the constitutive relation from limited observations under governing physical constraints. Results show that the proposed mobility-suppression function captures both weak- and strong-field behaviour and substantially improves chloride-profile prediction, with the RMSE at 60 V reduced by approximately 70% compared with the constant-mobility PNP model. At low voltages, field non-uniformity and mobility attenuation remain mild but persistent, whereas at higher voltages they relax more rapidly and the suppression decays faster with time. For the material and conditions tested, an applied-voltage range of approximately 30–50 V provides a practical starting window for approaching peak chloride flux while avoiding pronounced strong-field mobility suppression. These findings improve the physical basis for interpreting electromigration-based tests and electrochemical treatments, and provide further insight into durability-related transport analysis under strong electric fields.