Dynamic features and combustion state identification of burner flames in a 660 MW coal-fired boiler

The combustion state of burners in coal-fired boilers is crucial for furnace efficiency and safety. However, relying solely on static flame features limits the ability to capture flame dynamics and accurately assess combustion stability. To address these limitations, an improved Non-Smooth Data optical flow method is proposed for extracting flame dynamics. This method performs multi-scale estimation via an image pyramid, capturing motion features and reducing errors from limited spatial resolution. A Lorentzian robust function and noise-adaptive parameter are introduced to suppress disturbances from illumination changes, blurred boundaries and weak textures. An adaptive regularisation strategy is proposed to further enhance performance under varying load conditions. The proposed method is experimentally validated on a 660Â MW opposed-fired boiler. The results show that under stable combustion, its fluctuation range remains within a narrow band regardless of load level, indicating good consistency. However, when mismatches occur in the regulation of air and coal feed rates, the flame structure becomes significantly disordered and the optical flow field shows intense disturbances. Therefore, the proposed dynamic features can sensitively and effectively reflect abnormal combustion conditions, providing a reliable set of quantitative indicators and methodological support for real-time monitoring and stability diagnostics of burner flame behaviour in coal-fired boilers.