Corrosion Rate Prediction for Underground Gas Pipelines Using a Levenberg-Marquardt Artificial Neural Network (ANN)

This study addresses the challenge of accurately predicting corrosion rates and estimating the remaining life of underground gas pipelines, which is complicated by the complex interaction of physical factors and environmental conditions. Traditional models are inadequate in capturing these variables, leading to less reliable predictions, which this study aims to address by developing a more accurate and optimized artificial neural network (ANN) model. This study focuses on predicting corrosion rates and estimating the remaining life of underground gas pipelines using ANNs implemented in MATLAB. It incorporates both physical factors, such as maximum corrosion depth and pipe thickness, and environmental variables such as moisture, soil resistivity, and chloride concentration. The analysis identified corrosion depth and wall thickness as significant contributors, influencing material integrity by 20% and 16%, respectively. The optimal ANN model, with a Levenberg-Marquardt structure and one hidden layer of 10 neurons, achieved superior accuracy, with an MSE of 0.038 and R² of 0.9998. The study addresses the challenge of accurately predicting corrosion rates and remaining life in underground gas pipelines by developing an optimised ANN model. Its contribution lies in creating a highly accurate prediction tool that outperforms traditional models and enables more informed decisions for pipeline maintenance and safety