A synthesis of classical and adaptive control

In this paper we describe a new adaptive control algorithm which is specifically designed for for systems which are difficult to control. The models that we consider may exhibit nontrivial time variations; have significant delays; have nonminimum phase characteristics; and may be subject to measurement noise and plant disturbances. Our goal is to obtain robust control laws, and to simultaneously obtain good transient response and small steady state error. A two step procedure is taken. First, we design a linear controller which is based on some a priori knowledge of the plant. This controller incorporates predictions of future outputs which are obtained from online parameter estimates. It is argued that this makes the design of a robust controller easier than when predictions are not used. Secondly, the control law feeds back filtered estimates of the disturbance process, which are also obtained from the parameter estimator online. An optimal design technique is used to design an appropriate filter, thereby tuning the disturbance response of the closed loop system. A robotic welder at the Construction Engineering Research Laboratory in Champaign, Illinois is currently being used as a test bed for these theoretical results. Observations are described and future work is discussed.<>