Base drag estimation in suddenly expanded supersonic flows using backpropagation genetic and recurrent neural networks

Abstract

In recent years, base pressure management has gained a lot of industrial importance due to its applications in missiles and projectiles. For certain aerodynamic vehicles, the base pressure becomes a critical factor in regulating the base drag. That prompted the current work to develop input–output relationships for a suddenly expanded flow process using experiments and neural network-based forward and reverse mapping. The objective of forward mapping (FM) is to predict the responses, namely base pressure (β), base pressure with cavity (β_cav), and base pressure with rib (β_rib), for a known combination of flow and geometric parameters, namely Mach number (M), nozzle pressure ratio (η), area ratio (α), and length to diameter ratio (ψ). On the other hand, an effort is made to decide the optimal set of flow and geometric parameters for achieving the desired base pressure by reverse mapping (RM). Neural network-controlled backpropagation and recurrent and genetic algorithms have been employed to carry out the forward and reverse mapping trials. A batch mode of training was employed to conduct a parametric study for adjusting and optimizing the neural network parameters. Due to the requirement of massive data for batch mode training, the data required for training was achieved using the response equations developed through response surface methodology. Further, the forecasting performances of the neural network algorithms are compared with the regression models (FM) and among themselves (RM) through random test cases. The findings indicate that all evolved neural network (NN) models could make accurate predictions in both forward and reverse mappings. The results obtained would help aerodynamic engineers control various parameters and their values that affect base drag.