Development of a fast aerodynamic tool using meta-modeling techniques

SUMMARY There is a need for a wave drag method that combines the speed of handbook methods with the accuracy of computational methods. Especially determining the onset of wave drag as well as the initial drag rise is important in initial design stages. Meta-models allow this by capturing the trends present in previously computed data, providing an accurate and fast representation. In this report, it is investigated what gains can be achieved by applying the meta-modeling method GT-Approx to the aerodynamic tool MSES. The total drag calculated by MSES for a supercritical airfoil was verified using wind tunnel experiments. It was found that aerodynamic characteristics and pressure distributions are accurate up until M = 0.76. GT-Approx was critically evaluated to determine what accuracies are attainable, and how the resolution of the input data influences these accuracies. Initially the fitting capabilities of GT-Approx are tested. With a maximum error of 0.11% the fit is accurate. The errors increase when GT-Approx is used to predict values for data points that are not present in the data set. Especially in the case of extrapolation, the errors grow exponentially. If GT-Approx is used to predict values in between data-points, thus performing an interpolation, the results are better, with average errors of 2 drag counts. However, when predicting difficult regions, such as the drag coefficient near the drag rise, the errors increase. The prediction errors for high M are large, reaching 6.1 drag counts for c d w and 5.9 drag counts for c d v. In an effort to reduce these errors, the influences of resolution increases are investigated. Upon increasing the resolutions, the errors decrease. Especially increasing the M-resolution and c l-resolution yields large improvements. The average error for the higher Mach numbers reduces from 6.0 drag counts to 2.0 drag counts if the resolution is quadrupled. The average errors also reduce significantly, as shown in Table 1.