Modeling of PEO Coatings by Coupling an Artificial Neural Network and Taguchi Design of Experiment

An artificial neural network (ANN) was developed to predict the corrosion resistance of MgO coatings produced by the plasma electrolyte oxidation process on ZX504 alloy, with experimental data from the Taguchi method for training and performance evaluations. Process variables, i.e., chemical composition, current density, frequency, and duty cycle, were considered as inputs; and the corrosion resistance (measured by electrochemical impedance spectroscopy technique) as output data. According to the simulation results in the training stage, a high correlation coefficient was achieved between predicted and measured values (similar to 0.99). Using this well-trained ANN model, the proposed ANN model could predict corrosion resistance with a mean square error of approximately 0.0085 made from the test dataset. Examination of the surface morphology suggested a correlation between the microstructure and corrosion performance. The results showed that samples coated at lower Na3PO4, frequency, duty cycle, and higher KF and current density have the highest corrosion resistance.