Corrosion performance prediction of self-healing smart coatings on AZ31 magnesium alloys using feedforward neural network

Abstract

Investigating the self-healing behaviors of smart coatings incorporating microencapsulated linseed oil and honey as a corrosion protection method for AZ31 magnesium alloy is both time-consuming and expensive. To address these limitations, this paper proposes a feedforward neural network (FNN) to predict corrosion behavior using experimental data from the smart-coated alloy. The neural network was trained using corrosion data at a single stirring speed of 800 rpm and validated against three distinct datasets, including the original training data at 800 rpm, a separate dataset at the same speed, and additional experimental data obtained at 1100 rpm. The results show that FNN achieved strong predictive accuracy (less than 5% error) for corrosion potentials under matching 800 rpm conditions, though performance declined at 0% (10.09% error) and 40% honey concentrations (5.82% error). While effectively modeling fundamental potential–current density relationships at trained conditions (800 rpm), the accuracy of FNN diminished when tested at 1100 rpm, revealing sensitivity to stirring speed variations. However, the application of neural networks to predict corrosion behavior in smart-coated magnesium alloys shows significant promise. This approach improves the accuracy of understanding complex corrosion mechanisms in magnesium alloys, enabling the development.