Pitting corrosion behaviour modelling of stainless steel with support vector machines

Abstract

The knowledge of pitting corrosion behaviour of stainless steel is a critical factor in material science. In order to develop an automatic system to study pitting corrosion behaviour of this material, models based on support vector machines (SVMs) and k-nearest neighbour are presented in this work. The influence of the principal environmental factors involved in pitting corrosion, including chloride ion concentration, pH and temperature together with breakdown potential values obtained from polarisation tests, are analysed. Different feature selection techniques are applied (principal component analysis - PCA, linear discriminant analysis - LDA and Fisher criterion - FDR) as pre-processing stage. The results based on precision and accuracy indices prove that SVMs model using cubic polynomial kernel with LDA or FDR provides excellent classification performance. Therefore, this model becomes an effective strategy for modelling pitting corrosion and it may be considered as useful tool in the design stage of the structures.