| Résumé | Premixed gas turbine decreases exhaust emission but is vulnerable to combustion instability. This paper proposes a new method for early detection of combustion instability that damages combustors, thereby helping to operate energy production system in a more stable way. The proposed method is based on analysis of the spectral centroid and the spectral spread of dynamic pressure so that it enables early recognition of periodic waveform generation. To validate this method, transient dynamic pressure data from single and multi-mode instabilities, obtained from the H₂/CH₄/CO flames in a model gas turbine combustor, are used. The effectiveness of the method is compared with that of two conventional methods based on a root-mean-square dynamic pressure and a temporal kurtosis. Among these three methods, the method using spectral features shows the best performance in terms of speed, sensitivity, and accuracy in detecting combustion instability. However, the spectral method sometimes misclassifies unstable flames as having a stable status when multi-mode combustion instability occurs. Therefore, spectral features are modified by weighting the amplitude of dynamic pressure oscillation, which shows better performance as a precursor for multi-mode combustion instability detection. |
|---|
