Development of machine learning techniques in corrosion inhibition evaluation of 5-methyl-1 H-benzotriazole on N80 steel in acidic media

Abstract

In this study, three machine learning models of data analysis based on multiple linear regression (MLR), multiple nonlinear regression (MNR), and artificial neural network (ANN) has been proposed to model the behavior of 5methyl-1 H-benzotriazole (m-BTA) inhibitor on N80 steel in acidic media at different temperatures and inhibitor concentration. Based on experimental observation, the DC trend removal by Polynomial detrending revealed a suitable trend between electrochemical noise (ECN) measurement data and the results obtained from electrochemical impedance spectroscopy (EIS) and potentiodynamic polarization test at m-Factor equals to 3, 4, and 5. The inhibition efficiency (η) of 91% was achieved by using 0.001 M of m-BTA inhibitor. The corrosion rate will increase by increasing the temperature for the blank sample. Still, the m-BTA resulted in corrosion protection for N80 at high temperatures, and ECN resistance (Rn) increases with increased m-BTA concentration. The experimental tests revealed that temperature and m-factor are significant for corrosion inhibition of m-BTA; also, inhibitor concentration is an essential factor. Therefore, these three parameters were chosen as inputs for the Rn of the inhibitor. Removing unnecessary data from the original dataset causes relevant results compatible with experimental observations. The results of the reduced data set showed that R2 for the test of whole and reduced MLR model is 0.41 and 0.69, respectively, which shows that the reduced model is more suitable for modeling the training data than the whole model. The MNR model presents better accuracy than the MLR model, demonstrating that the relation between inputs and output does not follow a linear dependency. Several topologies of the ANN model were constructed, and the number of neurons was changed in one hidden layer. Finally, five neurons in the hidden layer with the Levenberg–Marquardt backpropagation (LMBP) algorithm were selected to develop the optimum ANN model, which has R2 and RMSE values of 0.99 and 17.41.