Prediction of corrosion lifetime of marine environment steel pipe piles based on physics-informed neural network

The corrosion of steel pipe piles in marine environments, particularly due to chloride ions, presents a significant challenge in marine engineering. Accurately predicting the service life of steel pipe piles at a low cost represents a critical technical challenge. This paper proposes a method based on the physics-informed neural network (PINN) to address this challenge. The method integrates the chloride ion diffusion model with the corrosion depth model to predict the service life of steel pipe piles. First, a chloride ion diffusion model is developed, and the relationship between the reduction in wall thickness and the degradation of mechanical properties is analyzed. Next, a multi-physics coupled mathematical model is formulated using Fick’s second law to predict the service life of steel pipe piles. Finally, by applying particle swarm optimization (PSO) to tune the PINN hyperparameters, the optimized model inferred an effective diffusion coefficient of D = 1.056 × 10⁻¹ mm²/s, deviating by only 4.3% from the experimental measurements. This study provides a low-cost, efficient solution for assessing the service life and informing maintenance decisions for steel pipe piles in marine environments, demonstrating substantial engineering applicability. A physics-informed neural network predicts steel corrosion in marine environments efficiently. The model links chloride diffusion and corrosion depth to assess structural lifespan. Experimental results confirm the model’s accuracy and potential for real-world applications.