OPTIMIZATION OF LEARNING ALGORITHMS IN THE PREDICTION OF PITTING CORROSION

Abstract

This work is part of a scientific research program whose objective is to prevent pitting corrosion of an open cooling circuit of a nuclear installation. Various corrosion inhibitors have been studied. The performances of pitting corrosion inhibition were discussed and compared on the basis of several criteria. The experimental data were collected in a large table and subjected to algorithms in order to construct models for predicting corrosion inhibition performance. We used four algorithms: Genetic Algorithm-Artificial Neural Network (GAANN); Least Squares-Support Vector Machine (LS-SVM), K Nearest Neighbors (KNN) and Regression Tree (RT). We optimized the data fraction reserved for learning and we sought to optimize the parameters specific to each algorithm. The efficiency of pitting inhibition increases in the following order: HCO3- < H2PO4- < CO32- < PO4-2 < PO4 3- < SiO3 2- < MoO4 2- < WO4 2-. Our results showed that the order of performance of the algorithms is: RT < KNN < LS-SVM < GA-ANN.

Keywords

• Genetic Algorithm-Artificial Neural Network (GA-ANN)
• Least squares support vector machine (LS-SVM)
• Regression tree (RT)
• K Nearest neighbor (KNN)
• Pitting potential (Epit)
• Inorganic inhibitors