Neural networks and correlation analysis to improve the corrosion prediction of SiO2-nanostructured patinated bronze in marine atmospheres

This article presents a new methodology that involves statistical tools and artificial neural network (ANN) modeling to predict the non-linear variation of the electrochemical impedance in two SiO2-nanostructured patinated quaternary bronzes in a marine atmosphere. The original experimental database provides the information on exposure time, [Cl-]; [SO2], relative humidity, precipitation level, wind speed, room temperature, the presence or absence of the nanocoating, corrosion potential, corrosion rate, frequency, and the real and imaginary parts of the impedance. All measurements were evaluated through descriptive statistical analysis and correlation matrix to find the variables that have the greatest linear influence on the imaginary impedance. For the artificial neural network modeling, the hyperbolic tangent sigmoid and radial basis in the hidden layer, and the linear transfer functions in the output layer were tested to find the best architecture using the experimental variables selected from the correlation matrix. The best-fitting training data set was obtained with 12 neurons in the input layer, 8 neurons in the hidden layer were used to achieve a coefficient of determination (R-2) above 0.99 in the 8 models tested, indicating the a good agreement between the simulations and the experimental data. The ANN models developed in this work allow us to successfully simulate the electrochemical behavior at low frequencies, which could represent a significant saving in future experimentation development.

Automated summary

This article presents a new methodology that uses statistical tools and artificial neural network (ANN) modeling to predict the non-linear variation of electrochemical impedance in two SiO2-nanostructured patinated quaternary bronzes in a marine atmosphere. The results show that with an appropriate neural network architecture, the electrochemical behavior at low frequencies can be successfully simulated.