Lean, premixed combustion has been aggressively pursued in recent years because it offers a practical approach for reducing emissions of nitrogen oxides (NOx) from gas turbines. However, lean premixed flames pose a greater risk of blowout. Studies on swirl and dump stabilized flames have shown that as a flame approaches blowout, distinctive precursors occur, such as pockets of localized extinctions and brief flame shape transitions to a thin ‘tornado’ configuration. For this study, both precursor types are detected using separate, dedicated optical sensors. Observations indicate that the tornado mode is part of the blowout process in a 127 mm long combustor and that a localized extinction precursor immediately precedes the tornado mode transition. Despite the causality, the statistics of tornado bursts and localized extinctions suggest a ‘memoryless’ Poisson process, where the occurrence of one event type does not influence the time until the next event of the same type. Localized extinctions have been used previously for blowout margin estimation, and are well suited for the purpose because the occurrence frequency increases with diminishing margin. However, the signals commonly used to detect localized extinction events are noisy due to the nature of the flame; thus, detection is prone to false alarms. Detecting tornado bursts, by comparison, is less ambiguous but such events occur too rarely for blowout mitigation applications. The shortcomings of both precursor detection methods can be addressed by combining observations of both precursor types in a meaningful manner. The presence of tornado bursts indicates that the flame is near blowout; this fact can be used to calibrate margin estimation routines based upon localized extinction. However, this approach would require two sensors since any one optical sensor cannot directly differentiate both precursor types. A single sensor approach can be developed whereby the causal relationship between the two precursor types is exploited. Local extinctions with longer duration times can potentially perturb the flame into an alternate flow configuration. The presence of these tornado ‘triggers’ manifest as an increase in the low frequency content of the chemiluminescence signal. A low pass filter with the appropriate cutoff frequency can differentiate between the tornado-triggering and benign, inconsequential localized extinctions. Therefore, the same signal that detects localized extinctions can be filtered to capably predict tornado mode shifts. This scheme will enable robust margin detection and minimize sensitivity to noise.

This content is only available via PDF.
You do not currently have access to this content.