Intelligent Flight Control System

Beginning in 1999, the NF-15B supported the Intelligent Flight Control System (IFCS) neural network project. This was oriented to developing a flight control system that could identify aircraft characteristics through the use of neural network technology in order to optimize performance and compensate for in-flight failures by automatically reconfiguring the flight control system. IFCS is an extension of the digital fly-by-wire flight control system and is intended to maintain positive aircraft con­trol under certain failure conditions that would normally lead to loss of control. IFCS would automatically vary engine thrust and reconfigure flight control surfaces to compensate for in-flight failures. This is accom­plished through the use of upgrades to the digital flight control system software that incorporate self-learning neural network technology. A
neural network that could train itself to analyze flight properties of an aircraft was developed, integrated into the NASA NF-15B, and evaluated in flight testing. The neural network "learns” aircraft flight characteris­tics in real time, using inputs from the aircraft sensors and from error corrections provided by the primary flight control computer. It uses this information to create different aircraft flight characteristic models. The neural network learns to recognize when the aircraft is in a stable flight condition. If one of the flight control surfaces becomes damaged or non­responsive, the IFCS detects this fault and changes the flight charac­teristic model for the aircraft. The neural network then drives the error between the reference model and the actual aircraft state to zero. Dryden test pilot Jim Smolka flew the first IFCS test mission on March 19, 1999, with test engineer Gerard Schkolnik in the rear cockpit.[1278]

The NF-15B IFCS test program provided the opportunity for a limited flight evaluation of a direct adaptive neural network-based flight control system.[1279] This effort was led by the Dryden Flight Research Center, with collaboration from the Ames Research Center, Boeing, the Institute for Scientific Research at West Virginia University, and the Georgia Institute of Technology.[1280] John Bosworth was the NASA Dryden IFCS chief engi­neer. Flight-testing of the direct adaptive neural network-based flight control system began in 2003 and evaluated the outputs of the neural network. The neural network had been pretrained using flight charac­teristics obtained for the F-15 S/MTD aircraft from wind tunnel testing. During this phase of testing, the neural network did not actually pro­vide any flight control inputs in-flight. The outputs of the neural network were run directly to instrumentation for data collection purposes only.

In 2005, a fully integrated direct adaptive neural-network-based flight control system demonstrated that it could continuously provide error corrections and measure the effects of these corrections in order to learn new flight models or adjust existing ones. To measure the aircraft state, the neural network took a large number of inputs from the roll, pitch, and yaw axes and the aircraft’s control surfaces. If differences were detected between the measured aircraft state and the flight model, the neural network adjusted the outputs from the primary flight computer
to bring the differences to zero before they were sent to the actuator con­trol electronics that moved the control surfaces.[1281] IFCS software evalu­ations with the NF-15B included aircraft handling qualities maneuvers, envelope boundary maneuvers, control surface excitations for real-time parameter identification that included pitch, roll, and yaw doublets, and neural network performance assessments.[1282] During NF-15B flight-test­ing, a simulated failure was introduced into the right horizontal stabi­lizer that simulated a frozen pitch control surface. Handling qualities were evaluated with and without neural network adaptation. The per­formance of the adaptation system was assessed in terms of its abil­ity to decouple roll and pitch response and reestablish good onboard model tracking. Flight-testing with the simulated stabilator failure and the adaptive neural network flight control system adaptation showed general improvement in pitch response. However, a tendency for pilot – induced roll oscillations was encountered.[1283]

Concurrent with NF-15B IFCS flight-testing, NASA Ames conducted a similar neutral network flight research program using a remotely con­trolled Experimental Air Vehicle (EAV) equipped with an Intelligent Flight Controller (IFC). Aerodynamically, the EAV was a one-quarter-scale model of the widely used Cessna 182 Skylane general aviation aircraft. The EAV was equipped with two electrical power supplies, one for the digital flight control system that incorporated the neural-network IFC capability and one for the avionics installation that included three video cameras to assist the pilots with situation awareness. Several pilots flew the EAV during the test program. Differences in individual pilot control techniques were found to have a noticeable effect on the performance of the Intelligent Flight Controller. Interestingly, IFCS flight-testing with the NF-15B aircraft uncovered many of the same issues related to the controller that the EAV program found. IFCS was determined to pro­vide increased stability margins in the presence of large destabilizing failures. The adaptive system provided better closed-loop behavior with

improved matching of the onboard reference model. However, the con­vergent properties of the controller were found to require improvement because continued maneuvering caused continued adaptation change. During ground simulator evaluation of the IFCS, a trained light-plane pilot was able to successfully land a heavily damaged large jet airliner despite the fact that he had no experience with such an aircraft. Test data from the IFCS program provided a basis for analysis and under­standing of neural network-based adaptive flight control system tech­nology as an option for implementation into future aircraft.[1284]

After a 35-year career, during which it had flown with McDonnell – Douglas, the Air Force, and NASA, the NF-15B was retired following its final flight, on January 30, 2009. During its 14 years at NASA Dryden, the aircraft had flown 251 times. The NF-15B will be on permanent dis­play with a group of other retired NASA research aircraft at Dryden.[1285]