Smart Corrosion Monitoring in AA2055 Using Hidden Markov Models and Electrochemical Noise Signal Processing

This work explores the application of Hidden Markov Models (HMMs) for the classification and reconstruction of corrosion mechanisms in the aerospace-grade aluminum alloy AA2055 from the signals obtained by electrochemical noise (EN) analysis. Using the PELT algorithm to segment the signal based on relevant changepoints, distinct corrosion states within the segments are isolated and identified, including general, localized, and mixed corrosion based on statistical signal features, which are used to create the probabilistic structure of HMMs through the initiation, transition, and emission matrices. This study utilized a dataset composed of five electrolyte groups, each containing ten EN signals with 1024 data points per signal, totaling 51,200 data points. The model demonstrates that even with variability in signal quality, meaningful reconstruction is achievable, especially when datasets include distinct transient behavior.