Early detection of thermoacoustic instability in an O2/CH4 single-injector rocket combustor using analysis of chaos and deep learning models

Abstract

Thermoacoustic instability (TAI) presents a critical challenge for lean-burning combustors and rocket engines. The early detection of instability is crucial, and to address this, a data-driven prediction framework has been established for TAI in a sub-scale rocket combustor with variable chamber length. Nonlinear combustion features are generated from time series of dynamic pressure using recurrence matrices. Deep learning models are then utilized to train these features and predict the proximity of impending TAI. The performance of the proposed method is investigated through cross-validations of 12 groups of hot-fire test datasets. Remarkably, the prediction performances are in good agreement with measured experimental data, with most instabilities being predicted dozens of milliseconds in advance. This capability paves the way for the early implementation of active control systems in full-scale combustors in the future. The prediction performances are also compared with state-of-the-art TAI prediction methods.