Abstract: Ion erosion is a key factor affecting the durability of concrete structures. Accurately predicting the ion diffusion within concrete is crucial for structural durability studies. A dual time-varying Fick's second law governing equation is constructed, which considers time-dependent characteristics of surface ion concentration and diffusion coefficient. A physics-informed neural networks (PINNs) framework is developed to solve the ion erosion diffusion. Bayesian optimization is used to optimize the PINNs structure, improving the solution efficiency. The PINNs model is used to solve the evolution of ion erosion diffusion and to evaluate the performance of concrete materials and service environments. The proposed model's accuracy is validated by COMSOL numerical simulations. Parameter analysis reveals the spatiotemporal evolution of ion erosion within concrete. The effects of different theoretical models on erosion diffusion solutions are compared. The study results show that, compared to the dual time-varying Fick's law, the traditional Fick's law underestimates the ion concentration on the outer side of the structure. It also overestimates the ion concentration on the inner side of the structure. Those findings demonstrate the necessity of considering the dual time-varying Fick's second law.