Corrosion Time Series Classification using the Continuous Wavelet Transform and MML Density Estimation

causes many failures in chemical process installations. These failures generate high costs, therefore an effective corrosion monitoring system obtrudes. This paper focuses on the classification of the most important corrosion processes: pitting, stress corrosion cracking (SCC) and general corrosion. The computations and algorithms involved in the classification of the corrosion time series are presented. A technique for trend removal of the time series is proposed. Features to extract the characteristics of the corrosion time series are designed. A genetic algorithm for the selection of features is described in which the correlations between features are exploited. The combination of the proposed techniques leads to a new high performing pattern recognition system for corrosion time series classification.