Artificial intelligence to predict inhibition performance of pitting corrosion

Abstract

This work aims to compare several algorithms for predicting the inhibition performance of localized corrosion. For this more than 400 electrochemical experiments were carried out in a corrosive solution containing an inorganic inhibitor. Pitting potential is used to indicate the performance of the inhibitor/oxidant mixture to prevent pitting corrosion. At the end of the electrochemical program a file containing all the experimental results has been prepared and submitted to several algorithms. Through a training phase each algorithm uses a set of experimental results to adjust its parameters and another set to predict the pitting potential starting from the properties and the chemical composition of the solution. The prediction performance of an algorithm is estimated by the difference between experimental pitting potential and the calculated one. The order of performance of the algorithms is: GA-ANN > LS-SVM > PSO-ANN > ANN >ANFIS > KNN > KBP > LDA.