Physics-informed neural network approach for identifying free chloride diffusion coefficient in alkali-activated mortars

To mitigate the high carbon emissions associated with conventional Portland cement production, alkali-activated cementitious materials (AACMs) have emerged as a promising sustainable and eco-friendly alternative. Chloride-induced steel corrosion represents a critical durability challenge hindering the practical application of AACMs. This study investigates the influence of key mixture parameters—water-to-binder ratio (w/b = 0.35 – 0.50), sodium silicate activator modulus (M_s = 1.0 – 1.8), alkali dosage (N = 5%–25%), and slag-to-fly ash ratio (S/F = 100/0 – 50/50)—on chloride transport in alkali-activated mortars using non-steady-state migration tests. Physics-informed neural networks (PINNs) are used to identify the free chloride diffusion coefficient (D_c), which is compared with the non-steady-state migration coefficient (D_nssm) to assess the implications for service life prediction. The results demonstrate that the Langmuir isotherm effectively characterizes the chloride binding capacity of alkali-activated mortars. Within the specific ranges, increases in w/b, M_s, and decrease in S/F result in elevated D_c. In contrast, the influence of N was found to be less pronounced. Neglecting nonlinear chloride binding in service life prediction models for reinforced concrete structures can lead to significant errors.