A technique, using recursive neural networks (RNN), has been developed for solving the six degree-of-freedom (6-DOF) equations of motion for an experimental radio controlled model (RCM) undergoing severe 6-DOF maneuvers. Experimental results and analyses indicate that the vehicle dynamics of the model during these severe maneuvers are dominated by forced unsteady fluid mechanics. Since neural networks have previously proven effective in modeling forced unsteady fluid mechanics, the approach taken was to extend these techniques to develop a neural network 6-DOF simulation of the vehicle dynamics based on the experimentally obtained data. Based on this paradigm, the only external inputs to the RNN were the time-varying control signals (Plane deflection angles and rates, Propeller RPM). The required outputs were the time-varying 6-DOF vehicle dynamics [u(t), v(t), w(t), p(t), q(t), rWl. The vehicle accelerations, Euler angles and trajectories were also calculated throughout the maneuver. Performance was judged by directly comparing the neural network predicted state variables and calculated trajectories to the experimentally recorded state variables and trajectories of the experimental model. The results showed that all variables (accelerations, velocities, angles and displacements) were accurately predicted throughout the entire maneuver. Further, all maneuvers could be accurately predicted over any time period during which the control inputs were a known function. The results also clearly showed that the RNN 6-DOF simulation generalized to severe maneuvers not used in the development process. Overall, the results * Johns Hopkins University Member AIAA Correspondence: David Taylor Model Basin, Code 503 Carderock Division, NSWC Bethesda, Maryland 20084-5000 t David Taylor Model Basin, Code 50 & 56 Carderock Division, NSWC Bethesda, Maryland 20084-5000 This paper is declared a work of the United States Government and is not subject to copyright protection in the United States. indicate that recursive neural networks provide an effective and accurate means for solving six degreeof-freedom equations of motion during severe maneuvers.
[1]
Thomas T. Huang,et al.
Experimental techniques to determine unsteady effects on a radio controlled model (RCM)
,
1995
.
[2]
William E. Faller,et al.
Unsteady Fluid Mechanics Applications of Neural Networks
,
1997
.
[3]
William E. Faller,et al.
Real-time prediction and control of three-dimensional unsteady separated flow fields using neural networks
,
1994
.
[4]
C. M. Ha,et al.
Neural networks approach to AIAA aircraft control design challenge
,
1991
.
[5]
William E. Faller,et al.
Quasi-linear neural networks: application to the prediction and control of unsteady aerodynamics
,
1993,
Defense, Security, and Sensing.
[6]
William E. Faller,et al.
Real-time prediction of three-dimensional dynamic reattachment using neural networks
,
1994
.
[7]
J. H. Jacobs,et al.
Use of Artificial Neural Networks for Buffet Environments
,
1994
.
[8]
Walter C. Merrill,et al.
Neural network application to aircraft control system design
,
1991
.
[9]
William E. Faller,et al.
Neural network prediction of three-dimensional unsteady separated flowfields
,
1993
.
[10]
Roger L. Simpson,et al.
Unsteady three-dimensional crossflow separation measurements on a prolate spheroid undergoing time-dependent maneuvers
,
1997
.
[11]
James E. Steck,et al.
Use of neural networks in control of high alpha maneuvers
,
1992
.
[12]
Robert F. Stengel,et al.
Identification of aerodynamic coefficients using computational neural networks
,
1992
.