DEEP LEARNING-ASSISTED ANALYTICAL STUDY OF INCOMPRESSIBLE UNSTEADY BOUNDARY LAYER BEHAVIOUR THROUGH DOUBLE-CHOKING SUPERSONIC EJECTORS

Abstract

This study presents a hybrid analytical–AI framework for predicting unsteady incompressible boundary-layer behaviour inside double-choking supersonic ejectors. Classical boundary-layer equations are first formulated to capture viscous diffusion, pressure-gradient forcing, and transient shear-layer dynamics influenced by primary and secondary choking. To overcome the limitations of analytical models in regions of strong nonlinear interaction, a deep-learning architecture integrating CNN, LSTM, and Physics-Informed Neural Networks (PINNs) is developed. Analytical solutions and high-fidelity CFD datasets are combined to train and validate the model. Results demonstrate strong agreement between hybrid predictions and CFD across velocity profiles, displacement and momentum thickness, wall-shear stress, and unsteady choking responses. The hybrid model significantly reduces error, reproduces separation zones, accurately predicts choking onset, and achieves up to 500× speed-up compared to CFD. Sensitivity analysis confirms robustness under varying Reynolds numbers and pressure gradients. The proposed framework offers a fast, accurate, and physically consistent tool for ejector boundary-layer prediction