Reference model supervisory loop for neural network based adaptive control of a flexible joint with hard nonlinearities

We propose an artificial neural network based adaptive controller for a positioning system with a flexible transmission element, taking into account hard nonlinearities in the motor and load models. A feedforward compensation module (ANN/sub FF/) learns the approximate inverse dynamics of the system and a feedback controller (ANN/sub FBK/) compensates for residual errors. The error at the output of a reference model, which defines the desired error dynamics, and the output of ANN/sub FBK/ are respectively used as the error signal for adaptation of ANN/sub FBK/ and ANN/sub FF/. The contribution of the paper is to propose a rule based supervisor for online adaptation of the parameters of the reference model to maintain stability of the system for large variations of load parameters. The controller is suitable for DSP and VLSI implementation and can be used to improve static and dynamic performance of electromechanical systems.