Camera based flame stability monitoring and control of multi-burner systems using deep learning based flame detection

The load flexible operation of industrial multi-fuel burners can lead to a destabilization of the flame, especially in low load conditions and when using a high share of biogenic (low rank) fuels. These flame instabilities can be circumvented up to a certain degree by online adapting fuel-, air- and swirl settings of the burner system. A necessary precondition for online burner adaptations is the availability of a quantitative real-time measurement of the flame stability, e.g. using camera systems. These measurement systems should be able to monitor multiple overlapping flames in a combustion chamber. Existing approaches for the flame stability measurement based on flame flicker analysis are rather empirical and give no defined value range and interpretation for a stable flame. For overlapping flames no flame stability measurement systems exist at all. In this paper we first present a new image processing based measurement system for the quantitative online flame stability monitoring. Using a deep learning based flame segmentation approach, the measurement system is able to monitor multiple overlapping flames. The new flame stability measurement has a geometric interpretation and can thus be directly applied to different burner-, fuel- and camera configurations. We then present a burner adaptation algorithm, which automatically adjusts the optimal burner settings to maximize the online measured flame stability. We demonstrate the new flame stability measurement and the burner adaptation algorithm at a 1 MW pilot-scale power plant for different fuels (charcoal, hard coal) using cameras in the visible and near-infrared spectral range with standard frame rates of 25 fps. •New camera based flame stability monitoring and control of multi-burner systems.•Using a deep learning based flame segmentation approach to monitor multiple overlapping flames.•New flame stability measurement has a geometric interpretation.•Burner adaptation algorithm to maximize the online measured flame stability.•Demonstration of new flame stability measurement and burner adaptation algorithm at a 1 MW pilot-scale power plant.