Long-term corrosion monitoring of carbon steels and environmental correlation analysis via the random forest method

Abstract

In this work, the atmospheric corrosion of carbon steels was monitored at six different sites (and hence, atmospheric conditions) using Fe/Cu-type atmospheric corrosion monitoring technology over a period of 12 months. After analyzing over 3 million data points, the sensor data were interpretable as the instantaneous corrosion rate, and the atmospheric “corrosivity” for each exposure environment showed highly dynamic changes from the C1 to CX level (according to the ISO 9223 standard). A random forest model was developed to predict the corrosion rate and investigate the impacts of ten “corrosive factors” in dynamic atmospheres. The results reveal rust layer, wind speed, rainfall rate, RH, and chloride concentration, played a significant role in the corrosion process.