Predicting Airspeed and Sideslip Angle using an Artificial Neural Network

This study has demonstrated the ability of an artificial neural network based method to predict airspeed and sideslip angle in both steady state and transient conditions in a single main rotor helicopter. Aircraft state and control positions were used as inputs to two multi-layer perceptrons, which separately predicted longitudinal and lateral airspeed. Simulation studies were used to investigate the potential of this method. Flight data from a Lynx Mk9 was used to validate the method in the low speed region: performance was seen to degrade in ground effect, and during transient manoeuvres, but overall performance fell within ±3.1 kn and ±10° of the true airspeed and sideslip values. In the high speed region a single multi-layer perceptron was used to predict airspeed, attaining an accuracy of ±2.6 kn. Taguchi methods were used in a post hoc analysis to identify those inputs that had the greatest contribution to the accuracy of the predictions.