Payload adaptive control of a flexible manipulator using neural networks

Abstract

Flexible manipulators provide significant advantages over the commonly-used rigid robots due to their lightweight properties, but an accurate control of these manipulators is more difficult to attain, and it is especially demanding in task executions involving changing payloads. This thesis addresses the problem of payload adaptive control of flexible manipulators. The nonlinear model describing the manipulator dynamics is completely derived and is then used for an accurate computer simulation of the flexible manipulator motions. Payload identification is implemented by using a novel neural network approach to identify distinct payload classes from tip deflection patterns which result from different payloads. The identification procedure is then used to select a controller which best meets the control objectives specifying hub speed and maximum tip deflection. Two distinct controller synthesis procedures, one using a pole-placement design and one employing a variable structure technique, are developed. The merits of payload adaptive control are shown by several simulation experiments.