Rapid prediction of the corrosion behaviour of coated biodegradable magnesium alloys using phase field simulation and machine learning

Surface protective coatings on magnesium alloys have been developed to control the corrosion rate of biomedical magnesium implants under mechano-chemical loadings. Quantifying the effect of coating’s microstructural features on the corrosion behaviour of magnesium alloys facilitates the innovative design of biodegradable magnesium implants from the surface to the bulk. The present work is devoted to exploring the applicability of deep learning methods for efficiently predicting the in vitro pitting corrosion behaviour of coated magnesium alloys. To this end, the proposed machining learning method employs different CNN models for predicting the corrosion curve and the evolution of corrosion interfaces. In the proposed deep learning method, phase field simulations with varying coating microstructures are used to generate the required corrosion datasets for training and validating the models. The method is applied to a PEO coated WE43 magnesium alloy to assess its feasibility based on in vitro experiments. Performance analysis shows that the multi-input CNN is superior to the single-input CNN in predicting the corrosion curve. The proposed encoder–decoder architecture can predict the evolution of corrosion interfaces with an average error about 1%. These results demonstrate that the proposed CNN models provide a promising alternative to conventional simulation methods for evaluating the protective performance of coatings.