Inspecting the epoxy/Nd2O3 composite coating electrochemical behavior using electrochemical impedance spectroscopy analysis with implementing exploratory data analysis and physics-guided machine learning prediction

This work investigates the impact of reinforcing various wt% of neodymium oxide (Nd2O3) varying from 1 to 10 wt%, on the electrochemical behavior of epoxy resin matrix through conducting the electrochemical impedance spectroscopy (EIS). Focusing on analyzing the solution resistance, charge transfer resistance (Rct), double layer capacitance and Warburg element. To explore the hidden seen and the complex nonlinear relation between the various wt% concentration and the electrochemical parameters, the exploratory data analysis and physics-guided machine learning (PGML) were used by integrating physics-guided spectral and shape descriptors extracted from Nyquist and Bode plots. The PGML was trained using this physical data by support vector regression (SVR) and Gaussian process regression (GPR) methods. to ensure model validation and robustness, the leave-one-out cross-validation was employed. Good agreement between PGML and EIS experimental data was accomplished and better prediction performance was observed by SVR compare to GPR. The sensitivity of Rct to the localized interfacial heterogeneities was confirmed by PGML results. The PGML showed high ability in capturing electrochemical structure-property correlation beyond analyzing the EIS with traditional equivalent circuit. Additionally, each wt% revealed to have unique effect on epoxy matrix, while a stability in the electrochemical behavior of nanocomposite was seen by the addition of 3 wt%-5 wt%.