A Monte Carlo Dual-RLS Scheme for Improving Torque Sensing without a Sensor of a Disturbance Observer for a CMG

Abstract

This article introduces a novel identification technique for the inverse model of a dynamical system by a dual-recursive least square (DRLS) algorithm to estimate the disturbance torque in the disturbance observer (DOB) configuration. Since the DOB uses the inverse model to estimate the disturbance, the accurate estimation of the inverse model affects the external torque estimation performance. To estimate the inverse model more accurately, Monte Carlo simulation is conducted for two RLS filters formed a back-to-back cascaded structure to identify both forward and inverse models simultaneously. Although the inverse model can be numerically driven by the identified forward model in the conventional way, we can have the improved accuracy of estimating the inverse model by dual-RLS filters. The effects on the external torque estimation accuracy by the proposed method are experimentally verified by evaluating the performance of estimating the disturbance torque in the control moment gyroscope (CMG) actuator.