Identification of corrosion nature using acoustic emission in representative scale samples

Monitoring corrosion is critically important across numerous applications, given its occurrence in key industrial sectors such as energy production, transportation, and civil engineering to prevent leaks and failures. Especially, determining the exact nature of corrosion phenomena could help planning specific maintenance operations, reducing costs and risks. Machine Learning methods have demonstrated capacities to discriminate corrosion phenomenona. An important step to create precise Machine Learning models is the quality of training datasets. With this goal in mind, this paper presents experiments performed on different metallic samples to create specific corrosion forms, respectively pitting and uniform corrosion to train a supervised model. This model is evaluated on 316L steel samples prone to endure both pitting and uniform corrosion. The usage of supervised learning is then discussed compared to other methods.