Effect of modeling approximations on the stability of autopilot controllers

The classical autopilot control law is extensively used for trimming rotorcraft models in comprehensive analysis codes. It consists of a simple control law that constructs a map relating the inputs and outputs of the system, based on a static approximation to its behavior. It is then easy to compute suitable filter time constants and control gains such that a closed loop controller will steer the system to its trimmed configuration with a desired performance. However, when this control law is used to steer complex rotorcraft models, such as those used in comprehensive analysis codes, stable behavior is only observed for judiciously chosen values of the controller parameters. Three major sources of error are responsible for the observed discrepancy: in the design of the controller, 1) the dynamic characteristics of the plant are ignored, 2) the non-linear behavior of the plant is not taken into account, and 3) the Jacobian of the system is assumed to be known exactly. This work focuses on the implications of these three assumptions on the behavior of the classical autopilot, by studying their effect through both numerical closed-loop experiments on a realistic UH-60 multibody rotor model (the plant), and eigenvalue analysis of the closed-loop characteristics of different reduced order models of the full plant.