DEVELOPMENT OF A MACHINE-LEARNING ENHANCED HIGH PERFORMANCE METHANE SENSING INSTRUMENT FOR FIELD APPLICATIONS

Abstract

Methane - a well-known potent greenhouse gas - has significantly contributed to globalwarming through their emissions from both natural sources, like wetlands, and human activities, including particularly the fossil fuel production sectors. The earlier to detect the emission sources, would allow a fast response to deploy the mitigation strategies to contain methane from releasing the environment. Therefore, making accurate and fast monitoring of methane emission is vastly crucial, particularly in the oil and gas industries where large-scale emissions are prevalent. However, the existing challenges on the field deployable sensing solutions still exist. They are either too costly involving human participation or with unstable or low sensitivity performance to provide an accurate and low false alarm sensing outcomes. Having this motivation, this thesis presents the development of a new cost effective, high performance and environmentally robust methane sensing instrument based on the NDIR (Nondispersive Infrared) sensing method and machine learning enhancement algorithm, which could be largely distributed to form a real-time methane monitor- ing network that can be strategically positioned for scalable and comprehensive area coverage, such as from facility-level to production basin or even regional level. Specifically, the thesis includes six chapters, starting from the background introduction to provide an brief overview of the existing methane emission issues and monitoring studies. Chapter two will provide a systematically review of the point sensor tech- nologies that can be used for creating real-time sensing network in the field to detect methane emission dynamics. Chapter Three will primarily describe the design of the circuit board and the overall setup of the device. After finalizing the design of a single device, Chapter Four will focus on the selection and training of the machine learning models used to process the data and mitigate environmental influences. Chapter Five will present the three validation methods we employed to verify the accuracy of the device’s readings: in-lab validation, open-area validation, and field validation.