Generalized Linear Modeling of Rotorcraft Acoustic Flight Test Data

The application of generalized linear modeling to rotorcraft acoustic emission data from an extensive flight test has been explored. The flight test data has 632 runs with at least 21 microphones measuring each run. The data was reduced to the maximum measured value for six acoustic metrics of interest, which are treated as the response variables in the modeling effort. Two sets of predictor variables were chosen, a dimensional set (true airspeed, gross weight, and flight path angle) and a non-dimensional set (advancing side tip Mach number, advance ratio, coefficient of weight, and wake skew ratio). It was shown that several of the predictor and response variables were not normally distributed, necessitating generalization of the linear models. A total of 19 different models were developed and analyzed, eight models were built from dimensional parameters and 11 from non-dimensional parameters. It was shown that the cubic version of the dimensional parameter model, which favors gross weight as a driving parameter, predicted the data set the best for all acoustic metrics. It was found that the response variables were from locations primarily directly beneath the vehicle, resulting in a predisposition towards modeling loading noise effects.