The identification of pitting and crevice corrosion spectra in electrochemical noise using an artificial neural network
Barton TF., Tuck DL., Wells DB.
An artificial neural network has been developed to identify the onset and classify the type of localized corrosion from electrochemical noise spectra. The multilayer feedforward (MLF) network was trained by classical back-propagation to identify corrosion from the characteristics of the initial current ramp. Using 50 training files and 39 test files taken from measurements on Type 304 stainless steel in a dilute chloride electrolyte, the network accurately detected and classified 96% of the data and reported no misclassifications. Experiments with high levels of adventitious noise superimposed on the original data have been carried out to examine the noise tolerance of the network.