Deep learning-based prediction of initiation jet momentum ratio in jet-induced oblique detonations

Oblique detonation, with its attributes of self-ignition, rapid heat release, and high thermal cycle efficiency, has garnered significant attention. It is crucial to explore methods that actively control the detonation at a shorter distance under less compressive conditions, and wall jets for active control stands out. However, when the jet momentum ratio excessively exceeds the critical value, it not only leads to resource wastage but may also compromise the outlet thrust performance. Thus, this paper presents a novel deep learning-based method for predicting the jet momentum ratio to ensure detonation with a sufficient margin while minimizing its value across various flight conditions. Firstly, the proposed methodology utilizes a deep neural network (DNN) to classify detonation status under different operating conditions. Building upon the concept of incremental learning, knowledge acquired by the classification network is repurposed to identify the transition point between detonation initiation failure and success. To address potential initiation failure issues in practical applications when seeking the critical jet momentum ratio, we elevate the threshold for the probability of successful initiation, defining such the ratio as the "robust jet momentum ratio". The framework developed in this study enables rapid identification of the robust jet momentum ratio within 0.07 s. To improve the model's inadequate learning of the discontinuous features associated with initiation/initiation failure transitions, data augmentation techniques are employed. The improved model demonstrates efficient determination under two-dimensional variables of flight altitude and Mach number. This study addresses a key challenge in jet prediction by devising an adaptive strategy for tuning jet momentum ratios according to varying flight conditions. This work bears significant engineering application value in the realm of hypersonic propulsion technology.