Technology Today

2010 Issue 2

Adaptive Flight Control Systems
Delivering more robust performance

The ability of a missile or an un­manned aerial vehicle (UAV) to complete its mission depends heavily on the quality of its flight control system. The quality of a traditional flight control system is rooted in the validity of the mathematical models used in its design, the fidelity of the information it receives in flight, and the health of its actuation de­vices. When the models are a good match to reality and the sensors and actuators are functioning as expected, uncertainty in the system is low and the vehicle behaves as de­signed and predicted. However, air vehicles do not always perform as their models would predict — due to battle damage, system faults, or aerodynamic uncertainties in the design models themselves — and as a result system performance degrades and mission effectiveness is reduced. However, by employing adaptive control algorithms that dynamically adjust to the changing conditions, performance can be maintained in the face of uncertainty. Raytheon has integrated adaptive algorithms into our missile and UAV flight control systems to deliver more robust performance.

While adaptive flight control has been an area of high interest in the controls commu­nity, it had not previously been applied to high performance missile systems or UAVs. Raytheon’s Adaptive Air Vehicle Technology (AAVT) strategic internal research and development effort has, for the first time, developed methods to utilize adaptive con­trol techniques in Raytheon missile and UAV autopilots. By partnering with academic researchers to investigate and refine various adaptive control methods, the AAVT re­search has been able to develop promising algorithms for real-world applications.

Consistent Performance for Uncertain or Degraded Systems

An adaptive flight control system uses one of two methods to maintain a consistent level of system performance in the presence of uncertainty and faults with minimal deg­radation. In the first, called indirect adaptive control, the adaptive controller monitors the difference between the measured system behavior and the expected system behavior in real time, estimates why those differences exist, and adjusts key control design param­eters based on those estimates to regain system performance.

The second method is direct adaptive control where the controller uses the per­ceived differences to compute an input control signal that directly drives those errors to zero without concern of why the differences exist. Whichever method is employed, the ability of an adaptive autopilot to provide consistent performance for uncertain or degraded systems reduces the need for high fidelity models and subsystem performance that is normally required for high-performance, robust autopilot design. By reducing the initial modeling effort and essentially doing more with less, adaptive control technol­ogy allows Raytheon to rapidly develop and deploy reliable flight control systems for ad­vanced missiles or UAVs at reduced cost.

Where a traditional flight control system would show degraded performance, the adaptive flight control systems developed maintain an expected level of system per­formance, as measured against a reference (nominal behavior) model. Many adap­tive systems either do not use a reference model, or use a simple reference model that is not consistent with the system dynamics. By adapting to the error between the desired response and the measured response of the vehicle in real time, the Raytheon adaptive controller creates an additional control signal that is used to augment that of a traditional robust autopilot. If the system dynamics match the represen­tative reference model, the contribution from the adap­tive controller will be zero. If, however, they do not match, the adaptive control signal will correct any differences and the combined control signal re­tains the desired performance.


Taking Algorithms from Academia to Missiles and UAVs

Throughout the lifespan of Raytheon’s AAVT research, various adaptive control techniques have been investigated. These include the neural network-based adaptive control developed at the Georgia Institute of Technology, L1 adaptive control developed at Virginia Polytechnic Institute and State University, the Retrospective Correction Filter (RCF) adaptive controller developed at the University of Michigan, and oth­ers. L1 and RCF are direct adaptive control algorithms that show many desirable char­acteristics and are the current algorithms of choice. Raytheon has partnered with the University of Michigan to assist with the application and evaluation of the RCF algo­rithm on flight vehicles.

Figure 1

While the academic research laid a foun­dation for the Raytheon development, a primary accomplishment of the AAVT re­search has been to develop the necessary modifications to these algorithms to suc­cessfully use them on missiles and UAVs. For example, one of the challenges in designing adaptive controllers for agile missiles and UAVs is the rapid response requirement, which dictates a high rate of adaptation in the con­troller. For missiles, system dynamics can change very quickly, so any adaptive action must react very rapidly. This typically is achieved by using a high adaptive gain value, but this can lead to high frequency control signals which are difficult for actuation systems to achieve. This is avoided in the L1 adaptive control algorithm through the use of a low pass filter on the adaptive control signal and a flexible companion model instead of a rigid reference model.

Adaptive Air vehicle Technology IRAD Adaptive Control Development Team

Figure 2

Another challenge is adaptively controlling the acceleration of a tail-controlled vehicle, where the actuation system is aft of the system’s center of gravity. To rotate the nose of the vehicle upward to produce a desired upward lift force, the control ac­tion must produce a downward force on the tail of the vehicle. This causes an initial downward motion to the vehicle center of gravity. A standard adaptive controller sees this initial ‘wrong-way’ effect and attempts to correct for it, resulting in an unstable response. One method to eliminate this problem is to instead control the angle of attack as measured by an air data system, but most missiles and proposed very small UAVs cannot sup­port an air data system due to weight and size restrictions. The Raytheon adaptive control­ler eliminates the wrong-way response problem by adapting to a careful combination of the measured acceleration and the measured angular rate, which are both available system out­puts. Overcoming these and other challenges to applying adaptive control to real-world systems have been among the achievements of the AAVT research.

Testing on Raytheon's Delta Wing Flight Control Test Bed

Because adaptive autopilots must operate in real-world systems, the AAVT research team developed a process for evaluation of algorithms that includes simulation and flight test. The algorithms are flown on Raytheon’s own Delta Wing advanced control system test bed. The Delta Wing is a low-cost UAV, developed in-house, complete with an internally developed avionics suite and processor. This vehicle demonstrated Raytheon’s first adaptive autopilot flight in 2007, and the team won the Raytheon Excellence in Engineering Technology award for this accomplishment in 2008.

Figure 3

Much is involved in the implementation and evaluation of adaptive autopilot algo­rithms for flight vehicles. After the adaptive flight control algorithms are developed and verified on simple examples, they are applied to a 6 degree-of-freedom (6DOF) simulation of the Delta Wing, where they are analyzed using Monte Carlo analysis. If the algorithm performs well in the 6DOF, algorithm performance is then verified during flight test of the Delta Wing flight control test bed.

A comparison of the pitch channel re­sponse of the Delta Wing 6DOF simulation to a step in the acceleration command is shown in the following figure. For this simulation, the pitch control authority was reduced 70 percent. The green line on the plot shows the response of the Delta Wing flying with a classical autopilot. The red line shows the response of the Delta Wing with an adaptive autopilot tracking the desired response from a reference model, shown in blue. Not only have these algorithms been applied to the Delta Wing UAV, they are being designed for and will soon be tested on the Cobra UAV, Raytheon’s own UAV test bed. Application to several advanced missiles is currently being pursued.

Throughout industry and academia, adap­tive control has been successfully applied to slowly varying industrial processes. It has also been applied to several aircraft and low-performance bombs. Through AAVT, Raytheon has pushed the state of the art in adaptive control by applying adap­tive algorithms to advanced air vehicles. Our systems are very challenging as they require extremely high performance and reliability, data sensing is often limited, and they may have either stable or unstable airframes. Furthermore, future systems are being developed that require the airframe to drastically morph its shape during flight. This poses very costly, if not insurmount­able, challenges to developing wind tunnel models for use in classical flight controller design. Additionally, future agile UAVs will require a high level of fault tolerance to meet airworthiness requirements. This will require on-board health monitoring and system identification, as well as a flight control system that can adapt rapidly to the detected changes. Finding solutions to these challenges is the future of adaptive control technology at Raytheon.

D. Brett Ridgely, Rick Hindm

Top of Page