Neural Network Autopilot System for a Mathematical Model of the Boeing 747

Abstract

Artificial neural networks can be defined as approximate mathematical models of the human brain's learning activities. In recent years neural networks have demonstrated abilities to perform autopilot and fault tolerant control tasks when applied to non-linear numerical aircraft simulations. Five on-line learning neural network autopilot systems, trained with the Standard and Extended Back-Progagation algorithms, were applied to a six degree-of-freedom non-linear simulation of a Boeing 747-200. The performance of the autopilots was compared based on their abilities to perform maneuvers at linear conditions and to adapt at non-linear conditions to restore steady state conditions. Linear maneuvers were performed by introducing reference values of altitude and speed, pitch angle, roll angle, or heading angle. The performance using the SBPA was satisfactory, but the EBPA performance was clearly superior throughout the entire range maneuvers while compensating for lightly damped phugoid and Dutch roll modes. Non-linear adaptation investigations were performed by exciting the non-linear terms in the equations of motion. The non-linear conditions were achieved in two ways: by simultaneously exciting pitch and roll rates with maximum elevator and aileron inputs, and the other by simultaneously exciting roll, pitch, and yaw rates with maximum elevator, aileron, and rudder inputs. The EBPA based controllers were able to regain steady state conditions for both non-linear tests with better transient performance than their SBPA counterparts. The SBPA showed only limited ability to adapt in cases where all three angular rates were excited. Artificial neural networks trained on-line using the Extended Back-Progagation algorithm are concluded to be better suited for autopilot systems for the 1/25 scale Boeing 747 based on their superior abilities to perform linear maneuvers and regain steady state conditions when at non-linear conditions.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Aug 04, 1998
Accession Number
ADA350857

Entities

People

  • Gerald C. Cottrill

Organizations

  • West Virginia University

Tags

Communities of Interest

  • Air Platforms
  • Space
  • Weapons Technologies

DTIC Thesaurus Topics

  • Air Force
  • Aircrafts
  • Algorithms
  • Automatic Pilots
  • Cargo Aircraft
  • Computational Science
  • Computer Simulations
  • Computers
  • Control Surfaces
  • Control Systems
  • Coordinate Systems
  • Equations Of Motion
  • Flight Control Systems
  • Mathematical Models
  • Models
  • Neural Networks
  • Transport Aircraft

Readers

  • Aerodynamics/Aeronautics.
  • Neural Network Machine Learning.
  • Robotics and Automation.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Machine Learning Algorithms
  • AI & ML - Neural Networks