FPGA implementation of learning rate supervisory loop for neural network based adaptive control of a flexible joint

Abstract

In this paper, we propose a VLSI implementation of a control strategy based on artificial neural networks (ANN) for a positioning system with a flexible transmission element, taking into account Coulomb friction for both motor and load, and using a variable learning rate for adaptation to parameter changes and to accelerate convergence. The control structure includes an ANN that approximates the inverse of the model and a reference model that defines the desired error dynamic. Adaptation of the learning rate of the ANN is used to reduce the sensitivity of the control structure to load and motor inertia variations. The paper presents a VLSI implementation of a supervisor that adapts the neural network learning rate on a Virtex2 Pro 2VP30 field programmable gate array (FPGA) from Xilinx. A pipelined implementation was used to speed-up the process. Simulation results highlight the performance of the controller and its response.