Predicting the Corrosion Rate of Oil and Gas Pipelines Using Neural Network

Abstract

Pipelines are popular means of fluid transportation. They have become the preferred medium of transporting hydrocarbon due to their cost-effectiveness, efficiency and safety. Therefore, in-use pipelines require adequate monitoring and maintenance for effective functioning. Pipeline inspection is a practice employed to prevent failures that could have significant consequences on their environments, aside from huge business losses. However, the fact that pipelines are mostly installed underground makes access and inspection challenges. Additionally, different subsurface materials have different chemical compositions and properties which could have a degrading reaction on the underlying pipeline material; thereby exposing the pipeline to failure risk. According to the pipeline failure record, one of the greatest causes of pipeline failure is corrosion. This paper developed a model for predicting the corrosion rate of oil and gas pipelines using neural networks. LevenbergMarquardt's (LM) backpropagation algorithm was used to optimize the training of the model for better predictive accuracy. The developed model was validated using MATLAB. Subsequently, the model was evaluated with an industrial dataset and was discovered to have an accuracy of 97%, this corresponds to improvements of 17.7% and 6.6% over the Obaseki analytical model and Abbas artificial neural network model respectively. The developed model has a root mean square error (RMSE) of 0.01421 and mean absolute error (MAE) of 0.00015, thus can accurately predict the corrosion rate of pipelines.