Study on Hysteresis of Combustion Oscillation in Continuously Variable Resonance Combustor using Deep Neural Network

Motohashi, Kazunori; Saito, Masanori; Tanabe, Mitsuaki

doi:10.2322/tastj.19.570

Cited by 1 publication

(2 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Motohashi et al shows that the activation of each neuron can be discussed using the chronos. 12) This method enables phase-wise descriptions. The combustion oscillation is mapped as a trajectory on the plane.…”

Section: Evaluation Data and Methodsmentioning

confidence: 99%

See 1 more Smart Citation

Study on Nonlinear Dynamics Using the State Space of Combustion Oscillation in a Rocket Combustor

ARAI,

IEMURA,

SAITO

et al. 2023

Trans. Japan Soc. Aero. S Sci.

Self Cite

View full text Add to dashboard Cite

Combustion oscillation in a rocket combustor was analyzed using a Variational Auto-Encoder (VAE). Dimension reductions were conducted on various physical quantities ( pA, _ q 0 , etc.) in the combustor. Orthogonal modes were obtained using proper orthogonal decomposition for the output of VAE. Pumping and damping regions in the combustor were identified using the distribution of the Rayleigh Index for each mode. Three characteristic modes were identified that supply energy to the oscillation. A phase plane was spanned by the latent variables obtained by the reductions. Activation of each of the characteristic six modes was mapped on the plane. The three characteristic modes were found to be activated at different phases. From the trajectory of the oscillation on the plane, the plane was confirmed to satisfy the requirements to be a state space. In the high-pressure amplitude phase, the pumping regions near the combustor wall were clarified.

show abstract

Section: Evaluation Data and Methodsmentioning

confidence: 99%

“…Recently, Deep Neural Network (DNN) functioning nonlinear transformations has been utilized widely in research fields. [11][12][13][14] Tanabe showed that the operation of dimension reduction using POD could be implemented in DNN. A method of mode decomposition using Deep Auto-Encoder (DAE) was also proposed.…”

Section: Introductionmentioning

confidence: 99%