Date of Award

December 2022

Document Type


Degree Name

Doctor of Philosophy (PhD)


Petroleum Engineering

First Advisor

Olusegun Tomomewo


Pipelines enable the largest volume of both intra and international transportation of oil and gas and play critical roles in the energy sufficiency of countries. The biggest drawback with the use of pipelines for oil and gas transportation is the problem of oil spills whenever the pipelines lose containment. The severity of the oil spill on the environment is a function of the volume of the spill and this is a function of the time taken to detect the leak and contain the spill from the pipeline. A single leak on the Enbridge pipeline spilled 3.3 million liters into the Kalamazoo river while a pipeline rupture in North Dakota which went undetected for 143 days spilled 29 million gallons into the environment.Several leak detection systems (LDS) have been developed with the capacity for rapid detection and localization of pipeline leaks, but the characteristics of these LDS limit their leak detection capability. Machine learning provides an opportunity to develop faster LDS, but it requires access to pipeline leak datasets that are proprietary in nature and not readily available. Current LDS have difficulty in detecting low-volume/low-pressure spills located far away from the inlet and outlet pressure sensors. Some reasons for this include the following, leak induced pressure variation generated by these leaks is dissipated before it gets to the inlet and outlet pressure sensors, another reason is that the LDS are designed for specific minimum detection levels which is a percentage of the flow volume of the pipeline, so when the leak falls below the LDS minimum detection value, the leak will not be detected. Perturbations generated by small volume leaks are often within the threshold values of the pipeline's normal operational envelop as such the LDS disregards these perturbations. These challenges have been responsible for pipeline leaks going on for weeks only to be detected by third-party persons in the vicinity of the leaks. This research has been able to develop a framework for the generation of pipeline datasets using the PIPESIM software and the RAND function in Python. The topological data of the pipeline right of way, the pipeline network design specification, and the fluid flow properties are the required information for this framework. With this information, leaks can be simulated at any point on the pipeline and the datasets generated. This framework will facilitate the generation of the One-class dataset for the pipeline which can be used for the development of LDS using machine learning. The research also developed a leak detection topology for detecting low-volume leaks. This topology comprises of the installation of a pressure sensor with remote data transmission capacity at the midpoint of the line. The sensor utilizes the exception-based transmission scheme where it only transmits when the new data differs from the existing data value. This will extend the battery life of the sensor. The installation of the sensor at the midpoint of the line was found to increase the sensitivity of the LDS to leak-induced pressure variations which were traditionally dissipated before getting to the Inlet/outlet sensors. The research also proposed the development of a Leak Detection as a Service (LDaaS) platform where the pressure data from the inlet and the midpoint sensors are collated and subjected to a specially developed leak detection algorithm for the detection of pipeline leaks. This leak detection topology will enable operators to detect low-volume/low-pressure leaks that would have been missed by the existing leak detection system and deploy the oil spill response plans quicker thus reducing the volume of oil spilled into the environment. It will also provide a platform for regulators to monitor the leak alerts as they are generated and enable them to evaluate the oil spill response plans of the operators.