Application and comparison of neural networks and optimization algorithms as a virtual angle of attack sensor

Angle of attack (AOA) measurement is an important part in flight control. AOA sensor failures caused several major accidents in aviation history (e.g. Flight 888T or Air France flight 447). In most cases one or two of three sensors fail due to e.g. ice freezing over and the flight computer chooses the faulty signal. A fourth sensor that works with a completely different principle could be compared to the remaining sensors and even used as replacement in case of total sensor loss. The presented method estimates an AOA to fit the data from thrust lever position, elevator position, airspeed sensor and acceleration sensors. For this purpose neural networks and optimization algorithms are compared with each other. The great advantage of this method is that it is applicable to nearly every flight computer and needs no prior knowledge of the airplane. Thus it could improve the security of flight control significantly.