Artificial neural network based shape optimization of supersonic ejectors in the critical flow regime

Abstract

Supersonic ejectors are widely used in energy, refrigeration, and aerospace propulsion applications. A supersonic ejector is a passive gasdynamic device consisting of a converging–diverging nozzle located in a varying area mixing duct where the secondary fluid is compressed by mixing with high momentum primary fluid. The entrainment ratio (ER, ratio of mass flow rates of the secondary to primary flow) is dependent on the mixing duct geometry, necessitating shape optimization. Complex gasdynamic interactions in the ejector are not fully understood, thus making physics-based or CFD modeling, a challenging exercise. In this paper, experimental data for ejector performance with varying geometries and operating parameters, viz. area ratio, primary flow Mach number, mixing duct length to height ratio (L/D), and stagnation pressure ratio is presented. The mixing length parameter is quantified using data derived from high-speed schlieren images. Improvement in ER to the extent of 20% is achieved by changing L/D to 15 for the AR of 3.33, with further increase in L/D resulting in a reduction in ER. Subsequently, computationally inexpensive model using ANN is developed to predict the ejector performance using the experimental database. The model predicts ER with an accuracy of 7%. Finally, the ANN model is used to optimize the ejector geometry for critical mode operation, which provides a 30% enhancement in entrainment ratio compared to the baseline geometry. • The effect of geometrical parameters on ejector performance is observed experimentally. • An optimum mixing duct height and length exist for maximum entrainment ratio (ER). • An ANN model is developed for performance prediction and shape optimization of air ejectors. • ANN predicts ER with an error of 7% compared to experimental ER. • ANN-optimized ejector offers a 30% higher entrainment ratio than the base ejector.