Use this URL to cite or link to this record in EThOS:
Title: Advanced flame monitoring and emission prediction through digital imaging and spectrometry
Author: Cugley, James
ISNI:       0000 0004 8511 0988
Awarding Body: University of Kent
Current Institution: University of Kent
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
This thesis describes the design, implementation and experimental evaluation of a prototype instrumentation system for burner condition monitoring and NOx emissions prediction on fossil-fuel-fired furnaces. A review of methodologies and technologies for burner condition monitoring and NOx emissions prediction is given, together with the discussions of existing problems and technical requirements in their applications. A technical strategy, incorporating digital imaging, UV-visible spectrum analysis and soft computing techniques, is proposed. Based on these techniques, a prototype flame imaging system is developed. The system consists mainly of an optical and fibre probe protected by water-air cooling jacket, a digital camera, a miniature spectrometer and a mini-motherboard with associated application software. Detailed system design, implementation, calibration and evaluation are reported. A number of flame characteristic parameters are extracted from flame images and spectral signals. Luminous and geometric parameters, temperature and oscillation frequency are collected through imaging, while flame radical information is collected by the spectrometer. These parameters are then used to construct a neural network model for the burner condition monitoring and NOx emission prediction. Extensive experimental work was conducted on a 120 MWth gas-fired heat recovery boiler to evaluate the performance of the prototype system and developed algorithms. Further tests were carried out on a 40 MWth coal-fired combustion test facility to investigate the production of NOx emissions and the burner performance. The results obtained demonstrate that an Artificial Neural Network using the above inputs has produced relative errors of around 3%, and maximum relative errors of 8% under real industrial conditions, even when predicting flame data from test conditions not disclosed to the network during the training procedure. This demonstrates that this off the shelf hardware with machine learning can be used as an online prediction method for NOx.
Supervisor: Lu, Gang Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: T Technology (General)