Software for Multidisciplinary Design Optimization of Truss-Braced Wing Aircraft with Deep Learning based Transonic Flutter Prediction Model
This study presents a new Python-based novel framework, in a distributed computing environment for multidisciplinary design optimization (MDO) called DELWARX. DELWARX also includes a transonic flutter analysis approach that is computationally very efficient, yet accurate enough for conceptual design and optimization studies. This transonic flutter analysis approach is designed for large aspect-ratio wings and attached flow. The framework employs particle swarm optimization with penalty functions for exploring optimal Transonic Truss Braced Wing (TTBW) aircraft design, similar to the Boeing 737-800 type of mission with a cruise Mach of 0.8, a range of 3115 n miles, and 162 passengers, with two different objective functions, the fuel weight and the maximum take-off gross weight, while satisfying all the required constraints. Proper memory management is applied to effectively address memory-related issues, which are often a limiting factor in distributed computing. The parallel implementation in MDO using 60 processors allowed a reduction in the wall-clock time by 96% which is around 24 times faster than the optimization using a single processor. The results include a comparison of the TTBW designs for the medium-range missions with and without the flutter constraint. Importantly, the framework achieves extremely low computation times due to its parallel optimization capability, retains all the previous functionalities of the previous Virginia Tech MDO framework, and replaces the previously employed linear flutter analysis with a more accurate nonlinear transonic flutter computation. These features of DELWARX are expected to facilitate a more accurate MDO study for innovative transport aircraft configurations operating in the transonic flight regime. High-fidelity CFD simulation is performed to verify the result obtained from extended Strip theory based aerodynamic analysis method. An approach is presented to develop a deep neural network (DNN)-based surrogate model for fast and accurate prediction of flutter constraints in multidisciplinary design optimization (MDO) of Transonic Truss Braced Wing (TTBW) aircraft in the transonic region. The integration of the surrogate model in the MDO framework shows lower computation times than the MDO with nonlinear flutter analysis. The developed surrogate models can predict the optimum design. The wall-clock time of the design analysis method was reduced by 1500 times as compared to the result implemented in the previous framework, DELWARX.