Resumo:
Corrosion is a global problem, which implies costs in industrialized countries of up to
4.5% of GDP, with either economic, but also social and environmental impacts. These
costs arise from waste due to maintenance caused by corrosion, which involves the
degradation of a metal in contact with an oxidizing solution. This indicates that
corrosion control should be promoted whenever possible. This study proposes a
methodology for electrochemical signal classification for a corrosion monitoring
system, in a system subject to the use of an inhibitor, through the passive approach of
electrochemical noise (EN) corrosion monitoring technique. The goal is to optimize the
parameters that promote optimal event classification in an EN corrosion sensor as part
of the methodology for a structural health monitoring (SHM) system. Due to its highly
dynamic and stochastic nature of the signal, this study and analysis of electrochemical
noise measurements (ENMs) consider numerical and graphical characteristics of two
corrosion systems in saline aqueous solution: carbon steel and stainless steel. These
experiments are repeated to collect data, which allows for generating graphs in time,
frequency, and time-frequency domains. From these graphs, certain characteristics are
extracted, such as mean, variance, standard deviation, skewness, kurtosis, gradient of
Spectral Power Density - PSD, PSD current cutoff frequency, low-frequency levels of
PSD, among others. These characteristics are assumed to have a good correlation with
physical data of the corrosion process. Next, a supervised machine learning system is
used to calibrate a model based on the training data. The complexity reduction selects
a limited set of parameters that, based on the test data, allow obtaining inference or
model accuracy in the upper quartile of the histogram of accumulated results with
accuracy of up to 100%.
