Thesis Defence: Upscaling Very Low-Resolution Chemiluminescence Images for Prediction of Thermoacoustic Source Term Using Machine Learning

Adam Atoom, supervised by Anas Chaaban and Sina Kheirkhah, will defend their thesis titled “Upscaling Very Low-Resolution Chemiluminescence Images for Prediction of Thermoacoustic Source Term Using Machine Learning” in partial fulfillment of the requirements for the degree of Master of Applied Science in Electrical Engineering.

An abstract for Adam Atoom’s thesis is included below.

Defences are open to all members of the campus community as well as the general public. Registration is not required for in person defences.

---

Abstract

Thermoacoustic oscillations in combustors require to be monitored and mitigated as they directly impact efficiency and safety of land-based power generation systems. Monitoring the thermoacoustic oscillations requires acquisition of high-resolution flame chemiluminescence images, which is not readily available in industrial settings due to the lack of optical access to the combustor. To mitigate the challenge, we propose using few sensors to produce low-resolution flame chemiluminescence images, which are then upscaled using machine learning to high-resolution flame chemiluminescence images. A multi-path convolutional upscaling neural network (NN) is developed and used in the present study to upscale the low-resolution flame chemiluminescence images to high-resolution flame chemiluminescence images. The effectiveness of the proposed NN is validated using image similarity metrics, and comparisons are made with existing image upscaling NNs in the literature. The performance of our NN is further assessed for predicting high-resolution spatial thermoacoustic source term. The proposed NN achieved more accurate upscaling and spatial thermoacoustic source term calculation compared to previous works in the literature. As a result, the present study demonstrates the feasibility of real-time detailed monitoring of thermoacoustic oscillations in industrial applications using few sensors.