Synergistic approach: Peridynamics and machine learning regression for efficient pitting corrosion simulation

Abstract

Corrosion-induced material deterioration poses a pervasive threat to structural integrity, necessitating an in-depth understanding of its intricate behaviors. Pitting corrosion, a critical concern in this context, accelerates the degradation of materials. The limitations of conventional models arise from their neglect of the subsurface electrode boundary layer dynamics during the dissolution process. In this study, we present a novel approach that combines Peridynamics (PD) diffusion framework with machine learning (ML) techniques to develop an efficient predictive model and computational efficiency. The proposed hybrid PD-ML model leverages the non-local effects inherent to Peridynamics and the pattern recognition capabilities of machine learning. It establishes an analytical connection between the concentration value at a specific material point and the concentrations exhibited by related constituents within its spatial horizon, considering the external mass flux applied. The adaptability of the model is achieved through the utilization of weighted regression coefficients, determined via multivariate linear regression. Validation against experiments and conventional PD model demonstrates the model's precision and efficiency using diverse micro-diffusivity scenarios. For 1D uniform and 2D pitting corrosion cases, our hybrid model yields precise concentration predictions while showcasing a remarkable improvement in computational speed compared to conventional approaches. Specifically, the hybrid model achieves an impressive speedup, approximately 4 times faster per time step and 2.5 times faster overall simulation. The study presents a promising tool for predicting corrosion-induced material deterioration in practical systems, offering accuracy, efficiency, and potential for broader applications.