A hybrid approach has been developed to enhance internal drag predictions because a precise accounting of aerodynamic forces is required for predicting mission performance in highly integrated systems. This approach is composed of wind tunnel experime...
A hybrid approach has been developed to enhance internal drag predictions because a precise accounting of aerodynamic forces is required for predicting mission performance in highly integrated systems. This approach is composed of wind tunnel experiments, numerical analysis, and Artificial Neural Networks (ANN). Through this framework, predicted exit flow properties were applied in a rigorous "thrust-drag bookkeeping" process. The ANN model is trained using a “reduced” dataset derived from the flow physics at the exit plane. The flow structure was obtained from numerical analysis which was validated against wind tunnel test results from the SNU RFL Direct-Connected Wind Tunnel facility.
The numerical model effectively captures key flow physics within a 13% error margin for total pressure recovery. To efficiently handle this high-dimensional data, Proper Orthogonal Decomposition (POD) was applied. With this method, the flow characteristics were extracted into dominant flow modes, reducing the dimensionality of the system while retaining over 99.9% of the flow energy. The final surrogate model reconstructed full-field exit profiles with high fidelity. Most notably, it predicted the integrated exit momentum for unseen validation cases with an error of less than 1.5% compared to the high-fidelity CFD results.
In conclusion, this hybrid approach allows for the decoupling of computational cost from query frequency, enabling rapid design space exploration. This hybrid framework offers a practical and powerful solution to the bottleneck of propulsion-airframe integration, representing a new paradigm for aerodynamic analysis and design.