Real-Time Adaptation of Mode Transition Controllers

Abstract

A real-time adaptation scheme is proposed for the online customization of mode transition controllers designed offline via blending local mode controllers. It consists of the desired transition trajectory model, the active plant model, and the mode transition controller. The active plant model, which incorporates local mode information, is initially trained offline to capture the desired transition trajectory and controls. Afterward, the active plant model is adapted online via structure and parameter learning to capture the input/output relationship of the nonlinear system to be controlled. Likewise, the blending gains portion of the mode transition controller is determined offline and is adapted online via structure and parameter learning to track the desired transition trajectory. The control sensitivity matrix and the one-step-ahead predicted output of the controlled system are used to develop the desired blending gains of the mode transition. The control sensitivity matrix and the predicted output are determined from the active plant model. The proposed adaptation scheme is illustrated for a hover-to-forward-flight mode transition control of a helicopter encountering parametric changes and wind disturbances.