Physics Informed Neural Network Model for Sustainable Marine Material Degradation Prediction Under Multiphysical Ocean Conditions

Abstract

Marine construction materials exposed to harsh ocean conditions gradually degrade as a result of series of processes, such as the effects of chloride-driven corrosive processes, temperature cycling, salinity, dissolved oxygen content, pH, hydrodynamic flow, mechanical loadings, creating serious problems related to sustainable marine infrastructure. In this research, Physics-Informed Neural Network (PINN) is proposed for prediction of marine materials degradation under various ocean conditions. Marine structural steel samples were subjected to controlled seawater conditions and analyzed using Scanning Electron Microscopy (SEM) and Electron Dispersive X-Ray Spectroscopy (EDS) in order to determine degradation indicators. Features were incorporated with environment parameters and material characteristics to create multi-physical feature matrix. PINN incorporates governing physical laws through the physics-based optimization method. Experimental validation produced R² = 0.956, RMSE = 0.083 mm, MAE = 0.061 mm which proves that degradation of marine engineering constructions can be predicted with high level of accuracy, reliability, and validity.