Dissertation Defence: Ethane Leak Detection from Surveillance Cameras in Petrochemical Industries

Junchi Bin, supervised by Dr. Zheng Liu, will defend their dissertation titled “Ethane Leak Detection from Surveillance Cameras in Petrochemical Industries” in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Electrical Engineering.

An abstract for Junchi Bin’s dissertation is included below.

Examinations are open to all members of the campus community as well as the general public. Please email zheng.liu@ubc.ca to receive the Zoom link for this defence.

---

ABSTRACT

Natural gas plays an important role not only in the global energy system but also in the manufacturing industries. Ethane, as the major flammable chemical compound in natural gas, is the primary input to the production of plastic and other industrial and consumer products. The growing demand for ethane from the global market requires Canadian petrochemical industries to refine millions of barrels per day for export. Ethane leaks from pipelines and facilities have severe consequences, such as economic losses, environmental impacts, and public safety. Infrared (IR) imaging is the prevalent hardware for achieving manual surveillance of ethane leaks. However, the manual process is laborious. The ethane leaks are challenging to distinguish in IR imaging through human inspection. Therefore, this research aims to develop an automatic framework to assist engineers in detecting ethane leaks from surveillance cameras.

First, a motion-aware ethane leak detection framework is proposed to detect ethane leaks from streaming videos of surveillance IR cameras. Specifically, the framework first applies a new tensor-based background subtraction (TBBS) to extract motion information from video frames. Then, an object detection framework, foreground fusion-based gas detection (FFBGD), is proposed to combine the motion information from TBBS and IR frames to accurately detect the ethane leaks. The experimental results indicate a significant improvement over the contemporary frameworks.

Second, the observation unveils that the visible (VI) camera can also help to visualize the ethane leaks. Integration of VI and IR cameras as multimodal imaging may also enhance ethane leak detection. Thus, at first, an image registration, \textit{i.e.} multi-objective optimization-based image registration (MOIR), is developed to align VI and IR images. Then, a vision Fourier transformer-based ethane detection (VFTED) is proposed to detect the ethane leaks by fusing the VI and IR images. The experimental results validate the feasibility of VFTED in improving ethane leak detection.

Third, a motion-aware multimodal ethane leak detection (MMELD) framework is proposed to integrate the advantages of these frameworks. The experimental results indicate that the proposed MMELD can further improve the detection accuracy and sensitivity in contemporary industrial frameworks.

In general, the outcome of this thesis contributes to the visual surveillance process in petrochemical industries. The proposed frameworks achieve precise detection and aid in the decision-making process of leak repairs. Meanwhile, the research conveys pioneer studies to improve the current ethane leak detection from surveillance cameras through two perspectives, i.e. motion information and multimodal imaging, which may also inspire academic researchers for future developments.