Abstract

The integrity of Oil pipeline infrastructure in Nigeria’s Petroleum Industry continues to be threatened by a mix of operational faults, equipment failures and external interfering factors, highlighting a real need for unique, intelligent and data driven monitoring solutions. In a bid to provide such a solution, this project developed and evaluated a Machine Learning based Binary decision support framework for early fault detection in Oil pipelines using data sourced from an operational flowline located in the Niger Delta region of Nigeria. The methodology adopted in carrying out the project was systematic, commencing with meticulous data preprocessing and finishing with rigorous model testing and evaluation. The result was a fault detection framework based on the XGBoost classifier algorithm trained on 263,053 points of data generated from operational field reports which yielded an AUC-ROC value of 0.9893 following a series of targeted algorithm optimization stages. Accuracy, Precision and Recall scores of 98.78%, 65.51% and 70.60% at a 0.80 decision threshold also served as further proof that the final model was effective and adequately adapted for reliable early fault detection in a real time deployment situation, while maintaining sufficiently low rates for false alarms. Additionally, the study proposed a low cost, integration pathway requiring minimal modification to existing systems, designed to seamlessly integrate the final binary model into existing pipeline monitoring frameworks. The final result was a self-improving intelligent decision support layer that meets the need for proactive and optimized pipeline integrity management in the Nigerian Oil and Gas industry.

Key words: Pipeline Integrity, Fault Detection, RTTM, Machine Learning, XGBoost, Niger Delta, SCADA systems.