Models for predicting corrosion inhibition efficiency of common drugs on steel surfaces: A rationalized comparison among methodologies

Abstract

Several anticorrosive treatments have been proposed over time to create protective layers to hinder the corrosion phenomenon. In recent years, organic molecules from plant extracts and expired drugs have been tested due to their potential corrosion inhibition properties. However, direct corrosion inhibition efficiency (IE%) evaluation requires costly reactants and a specific experimental setup. Quantitative-structure activity relationship (QSAR) proposes modeling IE% in terms of variables measured in previous experiments or determined by theoretical approaches. Computed descriptors, such as ionization energy (I), electronic affinity (A), or global hardness, were added to a database of physicochemical properties. This work compares several methodologies to obtain precise yet portable mathematical models for predicting corrosion inhibition efficiency. As an original approach from this research group, nonlinear autoregressive moving average with exogenous inputs (NARMAX), using forward regression with orthogonal least squares (FROLS), models were implemented as a robust method to get nonlinear portable models and to determine the most important variables impacting IE%. Contrastingly, ordinary least squares (OLS) methodology was employed with the novelty of applying power series expansions from the promoted FROLS variables for linear and polynomial regression with only one independent variable, which resulted in clearer graph visualization of trends and the ease of proposing thumb rules based on raw information. Finally, IBM Watson was also compared as a robust yet non-portable and highly parametrized alternative to conventional mathematical approaches, based on extra trees regressor (ETR). The models were compared using mean absolute percentage error (MAPE), mean-squared error (MSE), and root-mean-squared error (RMSE). Overall, models with fewer variables and up to second-order terms show improved performance. The main tendencies of IE%, drawn by inferences for 630 substances by second-order NARX, are analyzed. Also, the determinant role of the highest occupied molecular orbital energy was reported. Experimentalists can take advantage of a “cost-free” general approach that can obtain estimations for IE% values with errors of about 6 %, in particular the second-order NARX model.

Keywords

• Corrosion inhibition
• QSAR modeling
• HSAB theory
• quantum chemistry
• Artificial
• Intelligence