Experimental adaptive control of a hydraulic robot

Abstract

<p>An adaptive position controller is investigated to compensate for the effects of both the actuator and linkage dynamics in a four axis hydraulic robot. A simulation study was done, prior to the experimental implementation, in order to compare and contrast different adaptive control algorithms for this system. Implicit and explicit deterministic autoregressive moving average (DARMA) model based adaptive control algorithms were first simulated for hydraulic servo system. Simulation results indicated a satisfactory response. The results of this discrete-time, model-reference adaptive controller (MRAC) were compared with Craig's continuous-time, state-variable-based, adaptive controller. Simulation results showed that the position response between the two were comparable despite the discretization and unit delay incorporated in our model. The real-time implementation was then carried out using the DARMA model-based adaptive controllers. A model reference adaptive controller was first used for single axis testing. Experimental studies were conducted to see the effect of initial parameter estimates, choice of reference model, size of moves, and covariance modification on the position and control response as well as parameter convergence. The open loop data indicated a minimum phase plant, while closed-loop identification indicated a nonminimum phase behavior. This was due to the data sampling and computational latency of the control computer. Therefore, the MRAC could not be implemented for higher order models. Further in using first order model for MRAC, the controller had to be based on a model with no time delay. The pole assignment adaptive controller (PAAC) which does not require zero cancellation as in MRAC was chosen to overcome this difficulty. A second order DARMA model was used for identification and control. The position response of the PAAC was well damped with no overshoot after the initial transient period of identification. This alternative approach was used in further single axis testing of each axis. After successful single-axis tests on the individual axis of the robot, two, three and four-axis PAAC were implemented on the robot. In all the multiple-axis testing, well damped performance was achieved as desired after an initial transient period on all the axes.</p>