Generic neural flight control and autopilot system

TLDR

The paper proposes a generic neural flight control and autopilot system applicable across diverse vehicle classes. The system employs a neural controller that adapts flight control using reference models, receives guidance commands, and automatically schedules gains via frequency separation, and is validated through simulations on multiple aircraft types. Simulations show the generic neural system matches conventional performance and offers added resilience to damage or failures.

Abstract

This paper describes a generic neural flight control and autopilot system, which can be applied to a wide range of vehicle classes. A neural flight control system is used to provide adaptive flight control, without requiring extensive gain-scheduling or explicit system identification. The neural flight control system uses reference models to specify desired handling qualities, and can receive commands from a generic guidance system to provide outer-loop autopilot control. The generic guidance system performs automatic gain-scheduling using frequency separation, based upon the neural flight control system’s specified reference models. A variety of different aircraft were examined to ensure applicability to multiple vehicle classes including commercial transports, high performance military aircraft, and hypersonic concepts. Simulation results are presented for a mid-sized twinengine commercial jet transport concept, a modified F15 with moveable canards attached to the airframe, and a small single-engine uninhabited aerial vehicle hypersonic “waverider” concept. Results demonstrate that the generic neural flight control and autopilot system can achieve performance comparable to each aircraft’s respective conventional system, while providing additional potential for accommodating damage or failures.