An automatic pitting corrosion detection approach for 316L stainless steel

Abstract

Corrosion is considered a critical problem in many engineering structures and materials. Electrochemical techniques are the most popular techniques to study the corrosion behaviour of the materials. These techniques entail a surface study of the material to analyse its passive state. Therefore, the evaluation using these techniques requires visual interpretation steps which may lead to subjectivity in the results. In this work, different models based on artificial neural networks (ANNs), support vector machines (SVMs), classification tree (CT) and k-nearest neighbour (kNN) were presented to develop an automatic way to predict pitting corrosion behaviour of stainless steel in different environmental conditions. In addition, the influence of two different feature selection methods on the classification performance was considered. Receiver operating characteristics (ROC) space was applied to analyse the classification performance. The results, based on different statistic metrics considering 5-fold cross validation (93.1% of precision, 95.8% of sensitivity and 91.5% of accuracy), demonstrated the effectiveness of the proposed technique based on ANNs to predict corrosion behaviour by automatic way, not requiring the use of electrochemical techniques.