Short- and Long-Term Electrochemical Response Prediction of Ni-Al-Powder-Coated Steel with Machine Learning

Abstract

Steel is the most fundamental material used in structural system elements in the construction industry. It needs to be coated with materials that provide resistance to high temperatures, wear, and corrosion. Ni-Al powder is preferred in coatings because nickel increases corrosion resistance and aluminium forms an oxide layer to reduce oxidation. In the long term, the protective effect of coatings decreases, and corrosion resistance declines. In this study, a random forest model was evaluated using experimental data on the corrosion performance of A36 steel coated with Ni-Al powder for corrosion prevention, after exposure to a 3.5% NaCl solution for 1 h and 30 days for short- and long-term electrochemical response prediction. The impedance and phase angle characteristics, which represent the electrochemical response of coated and uncoated steel, have been predicted. In addition, the model’s reproducibility was investigated using the multi-seed (30 seeds) method to analyse the stability and consistency of the random forest model. The aim of this study was to develop a machine learning model that learns the frequency-dependent electrochemical impedance (Bode) response of graphene oxide-enriched Ni–Al coatings on steel, which reflects the corrosion-related electrochemical behaviour of the coating system, and to evaluate the model for predicting the impedance magnitude and phase angle of reference coatings over the investigated frequency range. The developed artificial intelligence model predicted the Bode response (impedance magnitude and phase angle) of coated and uncoated steel to NaCl solution after 1 h and 30 days as a function of frequency and coating type. The predicted impedance spectra reflected the deterioration of the corrosion protection performance of the Ni–Al coatings with increasing exposure time. The predicted EIS responses were subsequently used to assess changes in the corrosion-related electrochemical behaviour of the coatings over short- and long-term exposure. According to the findings, the random forest models can predict the frequency-dependent electrochemical response (impedance magnitude and phase angle) with high accuracy.