Detection of precursors of combustion instability using convolutional recurrent neural networks

Antony Cellier (1) , Corentin J. Lapeyre (1) , Gorkem Oztarlik (2) , Thierry Poinsot (2) , Thierry Schuller (2) , Laurent Selle (2)

1 CERFACS - Centre Européen de Recherche et de Formation Avancée en Calcul Scientifique
2 IMFT - Institut de mécanique des fluides de Toulouse

Many combustors are prone to Thermoacoustic Instabilities (TAI). Being able to avoid TAI is mandatory to efficiently operate a system without sacrificing neither performance nor safety. Based on Deep Learning techniques, and more specifically Convolutional Recurrent Neural Networks (CRNN)1, this study presents a tool able to detect and translate precursors of TAI in a swirled combustor for different fuel injection strategies. The tool is trained to use only time-series recorded by a few sensors in stable conditions to predict the proximity of unstable operating points on a mass flow rate / equivalence ratio operating map, offering a real-time information on the margin of the system versus TAI. This allows to change operating conditions, and detect the directions to avoid in order to remain in the stable domain.

Mots clés: Thermoacoustic instability; Instability precursors; Deep learning; Convolutional recurrent neural networks

Domaines: Mécanique des fluides [physics.class-ph]

Antony Cellier, Corentin J. Lapeyre, Gorkem Oztarlik, Thierry Poinsot, Thierry Schuller, et al.. Detection of precursors of combustion instability using convolutional recurrent neural networks. Combustion and Flame, 2021, 233, pp.111558. ⟨10.1016/j.combustflame.2021.111558⟩. ⟨hal-03382640⟩