Advanced aircraft are exploring increasingly novel concepts with significant departure from previous designs. This allows for unforeseen performance benefits, but also makes the process of Test and Evaluation (T&E) more expensive. In order to determine the aircraft performance and certify the aircraft for different maneuvers, aerodynamic databases have to be created. Traditionally, the creation of these databases has relied on wind tunnel and flight tests, which can quickly become expensive. However, with improvements in computational resources and increasing acceptance of Computational Fluid Dynamic (CFD) techniques, efforts are underway to allow use of simulation tools in creation of these databases (to reduce the burden on experiments). At the same time, complete reliance on simulations is also not possible at this stage. Hence, there is a need to merge information from CFD with experiment data in an informed manner, for the creation of multi-fidelity aerodynamic databases. Moreover, these databases have to incorporate estimates of uncertainty in the data, which is inevitable in the design and certification process. To address these concerns, we have been developing a toolset for creation of probabilistic aerodynamic databases that use multi-fidelity data and combine them using enhanced implementations of Gaussian Process Regression (GPR). An adaptive Design of Experiments (DOE) approach has also been developed to determine the next sampling locations for maximum benefit (informed by the application of interest). This paper describes the main features of the toolset and discusses recent enhancements for addressing non-Gaussian noise, alternate multi-fidelity formulation and source noise uncertainty estimation.
展开▼