Machine learning of unsteady transonic aerodynamics of a pitching truss-braced wing section

Abstract

A detailed analysis of transonic aerodynamics of a pitching conceptual Boeing Truss-Braced Wing (TBW) section has been carried out at various Mach numbers at a typical reduced frequency, k = 0.15. A new important flow feature of hysteresis loop cross-over has been discovered through the analysis of the $c_l$ hysteresis loops. Transonic dip has been located at M = 0.87, where the mean $c_l$ attains a minimum value. Hysteresis loop cross-over and transonic dip are important in the transonic flutter analysis of these wing sections. High fidelity database for the TBW section corresponding to different M in the range [0.7 - 0.9] has been generated by numerically solving the Navier-Stokes equations on a supercomputer at the NASA Advanced Supercomputing Division. A regularization-based machine learning (ML) methodology has been developed to predict the transonic aerodynamics of the pitching TBW section, using the training data as a subset of this high fidelity database. The ML model was then tested on test data as a subset of this database exclusive of the training data. Each ML model prediction is achieved well within a minute of computing time, as opposed to tens of hours of super-computing time required for each high fidelity CFD solution, thus making it a feasible tool for the design of a TBW, with the wing flutter in perspective. The TBW section flow was simulated corresponding to an altitude of 44,000 ft.