Corrosion prediction for MAO-coated magnesium using machine learning approach

Recently, magnesium alloys have attracted significant attention due to their ability to degrade within the human body, making them suitable for temporary biomedical implant applications. However, their rapid corrosion rates pose a challenge. To mitigate this issue, various techniques, such as micro-arc oxidation (MAO) coatings, have been employed. MAO effectively controls the degradation rate by forming a complex ceramic coating on the magnesium surface. Studies have shown that the corrosion behavior of ceramic-coated magnesium substrates is influenced by multiple factors, including electrolyte composition, process parameters, and other related variables. In this context, machine learning (ML) is a powerful approach that can be used to predict corrosion behavior based on such factors. This study presents the first application of ML techniques to predict the corrosion characteristics of MAO-coated magnesium implants based on potentiodynamic polarization (PDP) test results. Using data collected from the literature, data cleaning and transformation were performed before training the models. Multiple ML models were evaluated for forecasting corrosion current density and corrosion potential, with the Random Forest model achieving the best overall performance in terms of mean absolute percentage error (MAPE). Accuracy limitations were addressed by applying a logarithmic transformation to the current density values, which significantly improved prediction accuracy due to data range compression. Predicting log-transformed values directly yielded superior results compared to inverse-log transformations, particularly for current density, while predictions of corrosion potential remained stable across methods. Defining a common output range (10 −6 to 10 −7 A/cm²) further enhanced current density predictions but slightly reduced the accuracy for corrosion potential predictions. These findings highlight the importance of selecting prediction strategies based on specific performance priorities. Future work will integrate experimental validation and extend the approach to other coating systems. • Random Forest model best predicts PDP outcomes for MAO-coated Mg. • Log-transforming current density greatly enhances prediction accuracy. • Common range 10 −06 -10 −07 A/cm² improves current density predictions. • Corrosion potential predictions remain stable across transformations.