Detection of precursors of combustion instability using convolutional recurrent neural networks

摘要

Many combustors are prone to Thermoacoustic Instabilities (TAI). Being able to avoid TAI is mandatory toefficiently operate a system without sacrificing neither performance nor safety. Based on Deep Learningtechniques, and more specifically Convolutional Recurrent Neural Networks (CRNN) 1 , this study presentsa tool able to detect and translate precursors of TAI in a swirled combustor for different fuel injectionstrategies. The tool is trained to use only time-series recorded by a few sensors in stable conditions topredict 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 operatingconditions, and detect the directions to avoid in order to remain in the stable domain.