Subsystem-Based Learning Control of Hydraulically Driven Nonlinear Rotary Actuators with Unknown Input Backlash

Abstract

As hydraulic actuators are commonly meant to generate linear motions, the employment of mechanisms such as rack and pinion is necessary to convert such motions to rotation. However, these mechanisms present backlash nonlinearity that can degrade the control performance. Thus, this study proposes a subsystem-based learning (SSL) controller to operate a hydraulically driven manipulator (HDM) with a rotary hydraulic actuator (RHA) subjected to unknown input backlash and uncertainties. The proposed method necessitates neither knowledge of the backlash model nor the device that is subjected to backlash (gears or valves). The presented approach, which is based on virtual decomposition control (VDC), decomposes the entire system into rigid body and actuator subsystems, where local controllers are designed and stability analyses are performed. The novel way of incorporating radial basis function neural networks (RBFNNs) into VDC culminated in SSL controller, which allows us to tackle the uncertainties at each subsystem based on its characteristics, improving the estimation accuracy and increasing control performance. After ensuring the stability of each subsystem, the stability of the entire system is guaranteed by means of virtual stability and virtual power flow. Experimental results of implementing the designed controller on the commercial hydraulic manipulator are provided for performance evaluation. It is shown that the proposed method reduced the joint angle root-mean-square error from 0.2 degrees (corresponding to 1.5 cm at the crane tip position) to 0.095 degrees (equivalent to 0.6 cm at the tip), respectively, by tackling the system’s unknown backlash and uncertainties.