Pitting potential modeling using Bayesian neural networks

•Pitting potential values are predicted using artificial neural networks.•Chloride concentration, pH and temperature are considered for corrosion modeling.•Bayesian regularization is applied to select the optimal configuration of the model.•The model allows comparing pitting corrosion behavior in different environments.

Polarization tests have become a common technique to study pitting corrosion behavior of stainless steel. Based on these experimental tests, the pitting potential value may be determined at the potential at which current density reaches the value of 100 μA/cm2. The potential at this value of current density, associated to the breakdown potential, provides useful information about the likelihood of suffering pitting corrosion in the material under study: a more positive pitting potential means that the material can be expected to be more resistant to pitting corrosion. In this paper, a model based on Bayesian neural networks is presented in order to predict this potential value in different environmental conditions. In this way, it is possible to make comparison of corrosion resistance of the material under study in different environmental conditions. The results based on correlation coefficient and root mean square show the utility of this tool to predict potential values without requiring the use of polarization tests.