Search

Transonic flutter is a pacing item in transport aircraft design in that it is crucial to characterize this phenomenon for each aircraft to prevent catastrophic failure. Aerodynamic study of flows around airfoils is a canonical problem that entails both experimental and computational approaches. While the transonic flutter prediction can be more accurate with high-fidelity Computational Fluid Dynamics (CFD) methods than with unsteady potential flow methods, the computational cost is high. Therefore, computationally efficient methods for transonic flutter prediction continue to be of high interest to the aircraft design community. NASA Ames has developed a novel method that eliminates the need for expensive calculations of aerodynamics of wing flutter, which typically takes tens of hours on a supercomputer. Such calculations are now replaced by machine-learning-based closed form solutions that provide the solution almost instantaneously. The technology presents a new approach to predict the flow around pitching NACA00 series airfoils. NACA airfoils are generally symmetric, and thus they do not possess camber. However, the invention can readily extend to wings with camber. This novel data modeling approach is orders of magnitude faster than the traditional CFD approach of predicting aerodynamic effects of transonic pitching airfoils. The data model is based on a subset of unsteady CFD simulations that train the model. The trained model then resolves the pitching airfoil in time for any other set on the order of a second, as compared with a complete CFD simulation that typically takes 30 hours on a supercomputer. The data model is demonstrated in this invention for transonic flow corresponding to Mach number of 0.755 over pitching NACA00 series airfoils for a reduced frequency range typical of flutter, i.e., k lies in the range 0.02 - 0.25.

RAM-C interfaces with computational software to provide test logic and manage a unique process that implements three main bodies of theory: (a) aircraft system identification (SID), (b) design of experiment (DOE), and (c) CFD. SID defines any number of alternative estimation methods that can be used effectively under the RAM-C process (e.g., machine learning techniques, regression, neural nets, fuzzy modeling, etc.). DOE provides a statistically rigorous, sequential approach that defines the test points required for a given model complexity. Typical DOE test points are optimized to reduce either estimation error or prediction error. CFD provides a large range of fidelity for estimating aircraft aerodynamic responses. In initial implementations, NASA researchers “wrapped” RAM-C around OVERFLOW, a NASA-developed high-fidelity CFD flow solver. Alternative computational software requiring less time and computational resources could be also utilized. RAM-C generates reduced-order aerodynamic models of aircraft. The software process begins with the user entering a desired level of fidelity and a test configuration defined in terms appropriate for the computational code in use. One can think of the computational code (e.g., high-fidelity CFD flow solver) as the “test facility” with which RAM-C communicates with to guide the modeling process. RAM-C logic determines where data needs to be collected, when the mathematical model structure needs to increase in order, and when the models satisfy the desired level of fidelity. RAM-C is an efficient, statistically rigorous, automated testing process that only collects data required to identify models that achieve user-defined levels of fidelity – streamlining the modeling process and saving computational resources and time. At NASA, the same Rapid Aero Modeling (RAM) concept has also been applied to other “test facilities” (e.g., wind tunnel test facilities in lieu of CFD software).