Precursor detection of thermoacoustic instability using statistical complexity and artificial neural network

Abstract

Thermoacoustic instability (TAI) is a critical challenge for modern lean-burn combustion systems. This phenomenon is commonly undesired and should be avoided or suppressed to maintain high efficiency and structural safety. This paper proposes a methodology for categorizing combustion dynamical states and detecting the precursor of TAI. Statistical complexity of the acoustic pressure signals is captured by the complexity-entropy causality plane (CECP), and the precursor is detected by applying an artificial neural network (ANN) in CECP. The estimation provides an indicator of the proximity of the dynamical state to the onset of oscillatory instability and is well-validated in an annular combustor. It has been proven that ANN is more generalizable than the K-medoid clustering and can detect the transition ahead of those conventional methods indicated, namely, the temporal kurtosis and the root mean square. This study constitutes the demonstration of a novel framework that is particularly advantageous for detecting the onset of oscillatory instabilities of combustion systems.