Mathematical Regression and Artificial Neural Network for Prediction of Corrosion Inhibition Process of Steel in Acidic Media

Present work reports the results of the corrosion current density of mild steel in acidic solution as a function of acid concentration, inhibitor concentration, and temperature. Phenylthiourea was used as concentration a corrosion inhibitor. The work is focused on estimating the optimum mathematical equation and the ANN architecture. Corrosion rate (corrosion current density) was the response or the dependent variable, while inhibitor concentration (0.1, 0.75, and 1 g/l), acid concentrations (1, 3, and 5 M), and different temperatures (303, 313, 323, and 333 K) were the factors or the independent variables of the models. Five mathematical equations and five ANN architectures were suggested. Models were developed and estimated with the aid of a computer program. Linear, Linear – Logarithmic, Logarithmic – Linear, polynomial – individual effect, and polynomial – interaction effect model were estimated. Results show that the polynomial – interaction effect equation was able to accurately predict the measured data with acceptable correlation coefficient (R² = 0.9803). Linear Model, Multi-Layer Perceptrons, and Radial Basis Function were trained and tested. Multi-Layer Perceptrons with three inputs—multi hidden layers—three outputs (MLP 3:3-6-1:1) was the best ANN with significant performance (R² = 0.9529). Sensitivity analysis was carried out to study the effect of unavailability of each variable. Both mathematical and ANN analyses showed that the temperature has the most significant influence on the corrosion current density than the other two variables. The temperature individual effect coefficient was 840.6 as compared with 88.5 and − 170.5 for acid concentration and inhibitor concentration, respectively. Interaction effects between variables were also addressed.