Qin, Li (2023) NOx Prediction of a Combustion Process through Flame Imaging and Incremental Deep-Learning. Doctor of Philosophy (PhD) thesis, University of Kent,. (doi:10.22024/UniKent/01.02.103247) (Access to this publication is currently restricted. You may be able to access a copy if URLs are provided) (KAR id:103247)
PDF
Language: English Restricted to Repository staff only until September 2026.
This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
|
|
Contact us about this Publication
|
![]() |
Official URL: https://doi.org/10.22024/UniKent/01.02.103247 |
Abstract
This thesis describes the development of an intelligent methodology for predicting NOx emissions from hydrocarbon fuel combustion systems based on in-situ flame measurements and incremental deep learning models. The motivation, technical requirements and challenges of the subject area are addressed, followed by a comprehensive review of existing methodologies and technologies for monitoring and predicting NOx emissions. The key original contribution of this thesis is that a novel technical strategy that incorporates flame imaging, flame temperature mapping and advanced data-driven modelling is then proposed. Based on the proposed technical strategy, IDL (Incremental Deep Learning) models for predicting NOx emissions of an oxy-biomass combustion process are developed. Flame images acquired using a dedicated imaging system are used to calculate the temperature maps of the flame through two-colour pyrometry, which are then used as input datasets of the IDL models. The IDL model is combined with source DL (Deep Learning) and condition recognition models. The source DL model is initially constructed using flame data from 'seen' operation conditions and can be updated incrementally using flame data from 'unseen' conditions. A condition recognition model is used to identify whether the input data are from 'seen' or 'unseen' operation conditions (note that a 'seen' condition means a boiler operation condition having been used to train the IDL model, and a 'seen' condition means that having not been used to train the model). A new objective function that consists of regression loss and distillation loss is introduced in the IDL model to improve the prediction accuracy. Detailed model architectures, hyper-parameters determination and evaluation are reported. In addition, an IMCM (Incremental Multi-mode Condition Monitoring) model is developed for identifying 'seen' or 'unseen' multi-mode combustion conditions based on flame images. The IMCM model is built on the SMCM (Source Multi-mode Condition Monitoring) model, with updates incrementally using newly 'unseen' and a small portion of previously 'unseen' datasets. The loss balance weight, training epoch, and recognition probability threshold are carefully assessed for the optimal performance of the model. Moreover, the developed IDL and IMCM models are integrated, forming an IDL-IMCM model for predicting NOx emissions based solely on flame data. The IDL, IMCM and IDL-IMCM models are evaluated using test data from various oxy-biomass combustion conditions on an Oxy-fuel Combustion Test Facility. Results obtained from these models are presented and discussed. Compared to conventional CNNs, the developed IDL-IMCM model has an incremental learning ability that gives it great potential for online NOx prediction in practical combustion systems under variable operation conditions.
Item Type: | Thesis (Doctor of Philosophy (PhD)) |
---|---|
Thesis advisor: | Lu, Gang |
Thesis advisor: | Yan, Yong |
Thesis advisor: | Hossain, Moinul |
DOI/Identification number: | 10.22024/UniKent/01.02.103247 |
Subjects: | T Technology |
Divisions: | Divisions > Division of Computing, Engineering and Mathematical Sciences > School of Engineering and Digital Arts |
SWORD Depositor: | System Moodle |
Depositing User: | System Moodle |
Date Deposited: | 13 Oct 2023 07:42 UTC |
Last Modified: | 05 Nov 2024 13:09 UTC |
Resource URI: | https://kar.kent.ac.uk/id/eprint/103247 (The current URI for this page, for reference purposes) |
- Export to:
- RefWorks
- EPrints3 XML
- BibTeX
- CSV
- Depositors only (login required):