Artificial neural networks in flight control and flight management systems

Abstract

Different types of neural networks (NNs) are surveyed and their suitability as elements in flight control and flight management systems is analyzed. Advantages of neural networks over conventional digital avionic systems include speed (especially when implemented in special hardware, taking advantage of massive parallel processing), robustness, fault tolerance, and the ability to adapt to new situations by learning. An example shows how an artificial NN can be used as a gain adjuster in a stability augmentation system. A three-layer NN receives elevator commands and the sensed resulting longitudinal aircraft motion as input. The NN recognizes certain patterns in this response which are an indication that the gain is too high and that the control system is dangerously close to the stability boundary. Another example shows how a NN can solve a complex combinatorial problem which arises in search planning. It has similarities to the traveling salesman problem. >