Predicting pipeline corrosion in heterogeneous soils using numerical modelling and artificial neural networks

The influence of soil heterogeneity on the corrosion of underground metallic pipelines and the resulting evolution of localised corrosion patches were examined. A field-validated multiphysics numerical model coupled with random field realisations of the variables influencing corrosion was used in the investigation. The degree of saturation and saturated soil resistivity were considered as the most influential variables, and the numerical model outputs were used to train and validate an artificial neural network to predict the short-term and long-term corrosion rates given these input variables. The trained artificial neural network enabled rapid generation of corrosion profiles under various heterogeneous configurations of the input variables, implemented as random field realisations. Analysis revealed that the spatial variability of degree of saturation has a significant influence on the maximum corrosion patch size, depth, and frequency of occurrence. Saturated resistivity, while influencing the overall corrosion depth magnitudes, did not appear to influence the corrosion patch size configurations.