Notice and Invitation

Oral Defense of Doctoral Dissertation

The Volgenau School of Engineering, George Mason University



Zhenyi Ye

Bachelor of Science, George Mason University, 2018

Bachelor of Science, Southwest Jiaotong University, 2018





Gas Sensor Fusion for Intelligent Electronic Nose



Thursday, Nov. 18th, 2021, 10:00 AM-11:00 AM

ENGR 3507

All are invited to attend. RSVP required.



Committee

Dr. Qiliang Li, Chair

Dr. Dimitris E. Ioannou,

Dr. Brian L. Mark,

Dr. Jie Xu



Due to currently public health concerns as well as to comply with University requirements, RSVPs are required for this event. Please RSVP by Nov. 12th. To RSVP, please email [log in to unmask] to include your name, G# and GMU email. All attendees must complete Mason COVID Health✓™<https://www2.gmu.edu/mason-covid-health-check> and receive a “green light” status on the day of the event.



Abstract



The breakthrough in machine olfaction has long been desired by various applications from manufacturing to health care. Electronic nose (E-Nose) is a device based on gas sensor technology that mimics the human olfactory system. Compared to traditional aroma analysis instruments, E-Nose shows promising odor differentiation capability with the assistance of pattern recognition even without high specific receptors. Despite the recent advancements in E-Nose development, it remains challenging to achieve high-performance machine olfaction. To improve the differentiation capability of E-Nose, proper gas sensor fusion must be applied by making the most of sensor signals. This thesis introduces the gas sensor fusion for E-Nose to achieve precise gas differentiation qualitatively and quantitatively.



The design of the E-Nose system was studied, including hardware and control software. A modularized E-Nose design was proposed to separate the gas sensor interface with the controller and sensors for enhanced flexibility.  For qualitative aroma analysis, a dataset consisted of different odor categories was collected. Multiple methods, including traditional machine learning and various neural network architectures, were investigated and compared for their performance on the dataset. The results indicated improved odor differentiation accuracy for E-Nose with neural networks. A 1D convolutional neural network with ArcLoss was proposed to improve the feature quality and odor identification accuracy further. With more sensors involved during sensor fusion, E-Nose was also proved to produce reasonable quantitative predictions on blood glucose levels from human breath samples, which offered an insight into a non-invasive diabetes diagnosis method. In summary, this thesis presents a study on E-Nose sensor fusion from both system design and algorithm perspective, which is of great interest to E-Nose application development in the future.