Robust Controllers, Adaptive Systems

Robust flight control systems are designed specifically to perform well in the face of airframe, sensor, and actuator uncertainties and even failures. An early robust flight control system approach was the adaptive control system, a particular research objective of the Honeywell Corporation. This was in the days before airborne digital computers. The modest objective was to identify the airplane’s pitch natural frequency by periodic injection of small test pulses of elevator control. Pitch natural frequency variations reflected changes in both center of gravity location and dynamic pressure, or calibrated airspeed. Control system gain was lowered at the higher pitch natural frequencies to maintain system stability.

Modern applications of adaptive control make use of parameter identification, although test signals are still required to keep the parameter identification loop from going unstable. In a 1982 NASA workshop on restructurable controls, reasonably good results were reported for two adaptive schemes (Cunningham, in Montoya, 1983). Horizontal tail effectiveness Ms was identified on a Vought F-8 sufficiently well for autopilot gain scheduling through the flight envelope. Also, the flutter modes of a wind-tunnel model of a wing with stores (weapons) were identified by maximum likelihood methods.

The same NASA workshop brought a theoretical criticism of all adaptive systems by MIT professor Michael Athans. In his words:

Over two thousand papers have been written [on adaptive control] and a lot of excitement generated. You may have seen that people are giving courses to industry on how to make adaptive control practical. We have a recent MIT Ph. D. thesis [Rohrs, 1982] finished in November 1982 that Dr. Valvani and I supervised, which proved with a combination of analytical techniques and simulation results that all existing adaptive control algorithms are not worthwhile.

The algorithms may look excellent if you follow their theoretical assumptions, but in the presence of some persistent output disturbances and unmodeled high frequency dynamics all adaptive control algorithms considered become unstable with probability one.

Aside from coping with center-of-gravity and flight condition changes, robustness in control systems already exists in augmentation systems incorporating self-checking redun­dant digital computers. Robustness against sensor failures has also been demonstrated with redundant inertial sensors in skewed orientations. Failure of one or two sensors leaves the system fully operational. Failure of a single airspeed meter due to icing resulted in the losses of a General Dynamics B-58 Hustler and of an US/German X-31A research airplane. The automatic pilot gain-changing features interpreted the iced meter readings as low airspeed, requiring higher gains (communication from Dr. Peter Hamel).

Robustness against actuator failures, and especially against failures that result in control surfaces that go hard over against a stop and stay there, is another matter. The stirring example of Delta Airlines’ pilot McMahan who saved a Lockheed 1011 with one elevator against the up stop is told in Chapter 5, “Managing Control Forces.” System concepts for reconfiguring control systems to cope automatically with major failures are still in the early stages.

While waiting for the development of systems that are robust in the face of actuator hardovers, Thomas Cunningham suggests two straightforward aids for the human pilot. The position of each individual control surface should be measured and displayed in the cockpit. Captain McMahan did not know that the 1011 elevator was against its stop. Also, engine controllers should be designed to the higher bandwidths needed for differential thrust control of a crippled airplane.