N 9 3- 2
A Neural Based Intelligent Flight Control System for the NASA !=-15 Flight Research Aircraft James M. Umes Stephen E. Hoy Robert N. Ladage McDonnell Aircraft Company St. Louis, Missouri James Stewart NASA-Dryden Flight Research Facility Edwards, California A flight control concept that can identify aircraft stability properties and continually optimize the aircraft flying qualities has been developed by McDonnell Aircraft Company under a contract with the NASA-Dryden Flight Research Facility.
This flight concept, termed the Intelligent Flight Control System, utilizes Neural Network technology to identify the host aircraft stability and control properties during flight, and use this information to design on-line the control system feedback gains to provide continuous optimum flight response. This self-repairing capability (Figure 1) can provide high performance flight maneuvering response throughout large flight envelopes, such as needed for the National Aerospace Plane. Moreover, achieving this response early in the vehicle's development schedule will save cost.
On-Une Dee_ Cont_ _om_on Symm I_lyons: • H_h Neural Pedormance (Y, agnc,sl_s Netwo_ AJdtatne O=¢l • Non-Unear Law Moasuroment Conttot Ca,oa_l_ • High Agility of-.4JrU_ • Low Maintenance Deserting Co= ,Low Developmenl Cost
.! °.-
.._ Rep,tur L.b N_ I r Etc.
Figure 1. Self Designing Neural Flight System The Intelligent Flight Control System (Figure 2) incorporates an Aircraft Performance Model to provide the ideal system response. On-time measurements of Nrcmft _M Piio_ Performance Commands Ideal Model Response Specffi_U_ Effectorl p on........., o o . .
imcl !
Actuators SattlIO¢ll ) ) I !
I Aircraft )
_A _
Motion t Process Parameters Response U._ • I I I p i Aircraft !
TI
State I
§
I __ Matdx I Paramelers Matrix !
!
M0o_tq_ment p Figure 2. Intelligent Flight Control System the aircraft state parameters are determined by neural network models that relate aircraft stability coefficients (Figure 3), utilizing aircraft sensors such as Angle of Attack (AOA) as inputs to the networks. Thus, aircraft stability and control coefficients are continuously updated, and used in the control process to achieve the ideal desired response to pilot steering commands. The concept was designed to the NASA F-15 All Dstab Dailc Pdot Rdot CASNZ Beta I Dftail I Dail I Drud AOA! I Mach
, t I t
NN -i"
- :j "11" I
= ,t I[ , 1 C I C m C n Cy C z Figure 3. Neural Network Organization flight vehicle characteristics. Simulated response of the Intelligent Flight Control system to a pilot stick command is shown in Figure 4.
I ,- Model / i Pitch 8 Rate deg/sec 4 Stab 01] Com deg -4 r "-"£_"- ,,.. C .'i 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 Time - sec Rgure 4. Intelligent RigM Control System Response Mach 0.7 @ 20,000 Ft, GW-40,685 Ib, 1 inch Longitudinal Slick Step As a test of the concept, aircraft conditions representing a damaged wing was introduced into the problem, using the F-15 wind tunnel data for a 50% missing right wing (Figure 5). Neural Networks were developed to measure the damage, and tested using simulated time histories of the control system sensors as inputs to the networks.
F_ctom: (_) Determine F._nt ot Oam_ | (2) Determine Aircraft Stability I and Control Properties | (3) Revise Control Law I
I
Com_4ete in < 1 Second l In-Right ] Damage Detection Control 1_ Reconfiguration Figure 5. An Example Problem: Control of a Damaged Aircraft Figure 6 illustrates the aircraft response tLme history when the wing damage occurs.
Figure 6. F-15 Response: Right Wing 50% Missing The information from the Neural Networks will be used to quickly reconfigure the aircraft control surfaces and regain stable, controlled flight.
The Neural-based Self Designing Control Concept that is the basis of the Intelligent Flight Control system can be applied to future fighter and transport vehicles (Figure 7) to optimize engine and flight control performance.
--1 ,°0...,I-
S_19Otll Self.
aNI _ame Law
---i '=, 1--
Tra_lxxt High Speed CMI Figure 7. Neural-Based Self-DesigningFlight/Propulsion Control