Improvement of ENA - NOCS technique using Artificial Neural Networks approach for the detection of corrosion

摘要

We have conducted a feasibility study on the possible improvement of the output of Electrochemical Noise Measurements (ENA) conducted in a No Contact to Substrate (NOCS) setup identifying significant data mining methods for training Artificial Neural Networks (ANN) to finally assess the degradation degree of precorroded samples of unknown corrosion state. ENA experiments were performed on hot dip galvanized (HDG) and Al/Zn alloy coated steels, both artificially pre-corroded or exposed to outdoor conditions using a portable system consisting of a mini laptop, a 6 V2 digit digital multimeter and an in-house developed amplifier LPMH07. The technique was verified under laboratory conditions on realistic samples (precorroded HDG steel samples). The majority of experiments were conducted on samples having a well defined state of degradation. Nevertheless a closer look was taken at random samples to study the ability of the system to assess the corrosion state of different coatings. The chosen setup could prove its stability and reliability during these tests. To complete the feasibility study, a large number of parallel measurements were used to create a dataset from ENM experiments, which was used to train an ANN. The application of the evolved neural model on samples having an unknown level of corrosive degradation showed promising results. For the moment, the monitoring system "under development" allows to distinguish between 5 levels of degradation (very active/active/medium/passive/very passive) of galvanized and Al/Zn coated steel.