Accession Number : ADP006339
Title : Locally Linear Neural Networks for Aerospace Navigation Systems,
Corporate Author : DAYTON UNIV OH RESEARCH INST
Personal Author(s) : Gustafson, Steven C. ; Little, Gordon R.
Report Date : SEP 1991
Pagination or Media Count : 8
Abstract : Neural network software simulations for the representation and prediction of aircraft inertial navigation system (INS) data were developed. These simulations were evaluated using flight test data that sampled INS outputs at a standard rate for neural network testing and at half this rate for neural network training. The simulations used both locally linear neural networks and backpropagation-trained neural networks. Locally linear neural networks have several desirable properties for this application, including interpolation of the training data and representation of linear relationships. For the flight test data two milliradian testing accuracy was generally achieved with five successive and prior INS heading, pitch, and roll increments as inputs.
Descriptors : ACCURACY, AEROSPACE SYSTEMS, AIRCRAFT, COMPUTER PROGRAMS, COMPUTERIZED SIMULATION, EXPERIMENTAL DATA, FLIGHT TESTING, INERTIAL NAVIGATION, INTERPOLATION, LINEAR SYSTEMS, NAVIGATION, NETWORKS, NEURAL NETS, RATES, TEST AND EVALUATION, TRAINING.
Subject Categories : Air Navigation and Guidance
Computer Programming and Software
Distribution Statement : APPROVED FOR PUBLIC RELEASE