Enhancing Pipeline Reliability Analysis through Machine Learning: A Focus on Corrosion and Fluid Hammer Effects

Abstract

Natural gas, known for its cleanliness and cost-effectiveness, is transported across vast distances through pipelines. However, the safety concerns that arise from potential ruptures or leaks in these pipelines pose serious threats to the environment and human safety. This paper assesses the reliability of pipelines that have undergone corrosion, compounded by the fluid hammer effect observed in the liquefied gas flow. Machine learning models including support vector machines, linear discriminant analysis, random forest bagging, and Artificial Neural Networks have been meticulously crafted to forecast the safety status of pipelines, considering variables such as the pipe dimensions, material characteristics, fluid velocity, and flow rate. The design of the experiment methodology plays a pivotal role in calculating the pressure surge in pipelines corroded over time due to ongoing corrosion effects. The proposed machine learning models based on simulated data aim to predict the safety status of corroded pipelines with an accuracy rate of up to 97% in controlled environments. Integrating the proposed machine learning models for reliability analysis and pressure surge detection in corroded pipelines, in conjunction with the fluid hammer effect, offers an innovative approach to identifying risks and hazards.