A neural-network based inference engine for a general aviation pilot advisor
Tung Doan Nguyen, Donald T. Ward, Tung Doan Nguyen, Donald T. Ward · 35th Aerospace Sciences Meeting and Exhibit · 1997
An artificial neural network flight mode inference engine with the potential to considerably improve safety and streamline simulation training has been developed for possible use in general aviation advisory software. This first generation module is based on the backpropagation algorithm and was developed in an environment provided by commercially available neural network software. It is computationally feasible to train the parallel three-layer networks on available personal computers and to generate a training data set with a relatively small number of simulation flights on a fixed-base flight simulator. Comparison of this inference engine with a fuzzy logic based scheme gave similar results. Pilot evaluations using only two general aviation pilots showed that this first generation neural network inference engine detected certain pilot procedural errors more accurately than did the first generation fuzzy logic module. A more thorough evaluation, including flights with a larger number of representative pilots, of the two approaches is necessary before a definitive choice can be made. (Author)