Nonlinear mode decomposition of unsteady LOX/LNG flame imaging in an optically accessible rocket combustor

Kurzfassung

Combustion instabilities are characterized by nonlinear dynamics changing on short timescales. Common multivariable data analysis approaches like Dynamic Mode Decomposition linearise the underlying data source. The Koopman theory states that with a suitable transformation each nonlinear system can be linearly mapped with the Koopman operator. A variational autoencoder was trained in this work to approximate the Koopman operator along with the necessary transformation. This enabled analysis of nonlinear and time-varying effects in the radiation imaging of the flame in a LOX/LNG rocket combustor and the corresponding unsteady pressure data. In contrast to reconstructed DMD modes, the Koopman netwtwork provided a time resolved development of the underlying dynamics in the interaction between hydrodynamics and thermoacoustics in the development of combustion instabilities.