Optimal Control for Nonlinear Electromechanical Actuator

Abstract

The work done in this thesis focused on controller software for the Generic Short Stroke Actuator of Moog B.V. The actuator is controlled by the FCS controller, which makes use of a switched PID controller. This allows arbitrary control over the movement of the Generic Short Stroke Actuator. The FCS controller can switch between a position and force feedback loop. For various safety reasons the FCS controller has both force limits placed on it. If those limits are breached during operation, the FCS controller switches from position PID to force PID. Under certain conditions the even of a switch drives the actuator into a region of instability. The following thesis deals with finding the conditions for which the FCS controller drives the system into unstable regions. With the problems found, a solutions will be sought after from the field of Hybrid and Optimal Control theory. Three different approaches are investigated to solve the problems found: Dwell-time, Back Calculation Anti-Wind Up,Model Predictive Control. Dwell-time is a technique from the Hybrid Control theory field. A adequately found dwell-time guaranties the stability of any arbitrary switching system, like the FCS controller is. The Back Calculation Anti-wind up technique is a extension of PID control that deals with actuator saturation due to integrator wind up. Both of those techniques are improvements made to the FCS controller. Model Predictive Control comes form the filed of Optimal Control theory. Model predictive control tries to find optimal control input at each time step, taking into account any limits. In order to be able to design the above controllers, an accurate mathematical model was developed using fist principles. The model was optimised and validated against measured data from the system plant. As the plant is non-linear, to analyse it locally, the non-linear equations are linearised to establish a state space linear time-invariant system model. As the controllers are implemented on a computer, the model needed to be discretized to enable design of discrete controllers.