Search results

The purpose of this study is to confirm the idea that observing the electrochemical data of a steel polarized around its open circuit potential can provide insight into its performance against pitting corrosion. To confirm this idea a two-step work was carried out. The authors collected electrochemical data through experiments and exploited them through machine learning by building neural networks capable of predicting the behaviour of the steel against the pitting corrosion.

The electrochemical experiments consist in plotting voltammograms of the steel in chemical solutions of various degrees of corrosiveness. For each experiment, the authors observe how the open-circuit potential evolves over a period of 1 min, and following this, the authors observe the current evolution when they impose a potential scan that starts from the open-circuit potential. For each of these situations, the pitting potential Epit is noted. The authors then build different artificial neural networks, which after learning, can, by receiving electrochemical data, calculate a pitting potential Epit′. The performance of the neural networks is evaluated by the correlation of Epit and Epit′.

Through this work, different types of networks were compared. The results show that recurrent or convolutional networks can better capture the temporal nature of the input data.

The results of this work support the idea that the measurable electrochemical data around the free potential of a material can be correlated with its behaviour at more anodic potentials, particularly the initiation of pits.

To constitute input data, the authors carried out electrochemical experiments. The authors performed voltammetric scans in a very cathodic potential region. The authors constituted an experimental table where for each experiment we note the current values recorded at a low polarization range and the pitting potential observed in the anodic region. This study aims to concern carbon steel used in a nuclear installation. The properties of the chemical solutions are close to that of the cooling fluid used in the circuit.

In a previous study, this paper demonstrated the effectiveness of machine learning in predicting the localized corrosion resistance of a material by considering as input data the physicochemical properties of its environment (Boucherit et al., 2019). With the present study, the authors improve the results by considering as input data, cathodic currents. The reason of such an approach is to have input data that integrate both the surface state of the material and the physicochemical properties of its environment.

The experimental table was submitted to two neural networks, namely, a recurrent network and a convolution network. The convolution network gives better pitting potential predictions. Results also prove that the prediction by observing cathodic currents is better than that obtained by considering the physicochemical properties of the solution.

The originality of the study lies in the use of cathodic currents as input data. These data contain implicit information on both the chemical environment of the material and its surface condition. This approach appears to be more efficient than considering the chemical composition of the solution as input data. The objective of this study remains, at the same time, to seek the optimal neuronal architectures and the best input data.

This paper aims to predict the localized corrosion resistance by the application of artificial neural networks. It emphasizes the importance to take into account the relationships between the physical parameters before presenting them to the network.

The work was conducted in two phases. At the beginning, the authors executed an experimental program to measure pitting corrosion resistance of carbon steel in an aqueous environment. More than 900 electrochemical experiments were conducted in chemical solutions containing different concentrations of pitting agents, corrosion inhibitors and oxidant reagents. The obtained results were collected in a table where for a combination of the experimental parameters corresponds a pitting potential Epit obtained from the corresponding electrochemical experiment. In the second step, the authors used the experimental data to train different artificial neuron networks for predicting pitting potentials.

In this step, the authors considered the relationships that the chemical parameters are likely to have between them. Two types of relationships were taken into account: chemical equilibria which are controlled by the pH and the synergistic relationships that some corrosion inhibitors may have when they are in the presence of a chemical oxidant.

This comparative study shows that adjusting the input data by considering the physical relationships between them allows a better prediction of the pitting potential. The quality of the prediction, quantified by a regression factor, is qualitatively confirmed by a statistical distribution of the gap between experimental and calculated pitting potentials.