Prediction of the transient coolant jet released from the nose cone at supersonic flow via machine learning.

In this paper, the usage of a predictive surrogate model for the estimate of flow variables in the transient phase of coolant injection from the nose cone by combining the Long Short-Term Memory (LSTM) and Proper Orthogonal Decomposition (POD) technique. The velocity, pressure, and mass fraction of the counterflow jet is evaluated via this hybrid technique and the source of discrepancy of a predictive surrogate model with Full order model is explained in this study. The POD modes for the efficient prediction of the different flow variables are defined. The performance of the POD + LSTM for different ranges of training and test is evaluated and it is found that the performance of this hybrid technique is acceptable when 80% of the available data is training test. The predictive errors of coolant mass fraction and axial velocity is higher due to the complexity of the vortex in the recirculation region.