A comprehensive approach based on SVM to model pitting corrosion behaviour of EN 1.4404 stainless steel

Abstract

Potentiodynamic polarisation tests have been widely used to evaluate pitting corrosion behaviour of stainless steels. This technique involves two principal steps: the interpretation of the polarisation curve and the microscopic analysis of the material surface. Therefore, analysing the influence of all possible environmental conditions on pitting corrosion behaviour can be complex. To deal with this problem, automatic models based on Bayes classifier (BC) and support vector machines (SVMs) techniques are presented in this work. The principal environmental variables involved in pitting corrosion: chloride ion concentration, pH and temperature are considered. Based on the obtained results related to precision and accuracy terms, SVMs models using linear or radial basis function kernels outperform models based on Bayes theory. In addition, ROC analysis reveals that SVMs models using linear kernel is the best option providing maximum values of sensitivity and specificity. These results reflect the validity of the presented model to be applied in the automatic determination of the corrosion state of austenitic stainless steel as function of environmental conditions, not requiring the use of polarisation tests.