Explainable artificial intelligence-based insights into the corrosion behavior of WS2/AZ91 composites subjected to severe deformation conditions

Abstract

Magnesium (Mg) alloys have found potential applications in aeronautical, automotive, 3C industries, and the like owing to their good machinability, high specific strength, and low density. However, one of the main obstacles in impeding the Mg is weak corrosion resistance. Herein, the corrosion behavior of WS2/AZ91 composites and the effect of severe deformation through equal channel angular pressing was investigated experimentally and analytically via three-electrode system in a 3.5 wt% NaCl solution and data driven modelling. The experimental data of the current density and corrosion potentials of different composites at different deformation conditions was first correlated by Pearson, Spearman, and Kendall correlations. After that Bayesian surrogate Gaussian process (GP) assisted optimal neural network was developed to assess the corrosion behavior of different metal matrix composites at different deformation conditions. The correlation matrix showed that for different weight concentrations such as 0 wt %, 0.6 wt %, and 1 wt %, the Pearson correlation value becomes 0.77, 0.64, and 0.7, respectively. Similar to the Pearson correlation, the Kendall and Spearman correlations also showed relatively higher values for 0 wt % and 1 wt % compared to 0.6 wt % concentration. The proposed neural network model expressed a great accuracy in terms of correlation coefficient (R2 = 0.9668), mean absolute error (MAE = 0.0583), mean square error (MSE = 0.0405), and mean absolute percentage error (MAPE = 2.183). Although different concentrations and deformation conditions were included in the data, yet, the proposed DNN model was able to predict the current density data with a great accuracy. Finally, the explainable artificial intelligence was used to interpret the prediction of the developed model for different deformations and composite concentrations.