Flight Turbulence Level Classificator using a Multilayer Perceptron Network Trained with Flight Test Data
Keywords:
Artificial Neural Networks, Cross Validation, Flight Test, Flight Turbulence, Levenberg-Marquardt Backpropagation, Multilayer Perceptron, Turbulence Level ClassificationAbstract
This study presents the development of an artificial neural network (ANN) that classifies from flight data, the flight turbulence level encountered by an aircraft. The input data is divided into three different groups that contributes for turbulence level classification: Flight Condition, Aerodynamic Configuration and Turbulence Measurement. There are two main methods applied at turbulence measurement, Power Espectrum Density of aircraft vertical acceleration signal and Discrete Gust calculated from inertial and anemometric aircraft data source. The ANN model developed is a Multilayer Perceptron which was trained with Levenberg-Marquardt Backpropagation algorithm, using flight test data of a specific aircraft prototype. The flight test data used at learning process consists of both recorded parameters and flight test crew subjective flight turbulence level classification. The most precise model developed (within the sixteen models proposed and analyzed) were trained also with Cross Validation method due to lack of samples that represents all possible characteristics of the flight turbulence phenomenon.