The mass balancing of robotic manipulators has been shown to have favorable effects on the dynamic characteristics. In actual practice, however, since conventional manipulators have flexibility at their joints, the improved dynamic properties obtainable for rigid manipulators may be influenced by those joint flexibilities. Firstly this paper investigates the effects of the joint flexibility on the dynamic properties and the controlled performance of a balanced robotic manipulator with a PD controller. The natural frequency distribution and damping characteristics were investigated through frequency response analyses. To evaluate the dynamic performance a series of simulation studies of the open loop dynamics were made for various trajectories, operating velocities, and joint stiffnesses. These simulations were also carried out for the balanced manipulator with PD controller built-in inside motor control loop. through the diverse simulation, it was found that, at low speed, the joint flexibility nearly does not influence the performance of the balanced manipulator, but at high speed it tends to render the balanced manipulator susceptible to vibratory motion and yields large joint deformation error.
The lowered structural natural frequencies and reduced velocity realted terms such as coriollis and centrifugal term, through mass balancing, are supposed as the two main sources of the somewhat high vibrational tendency. This vibrational tendency of the flexible joint balanced manipulators in high speed motion limits the servo gain of the conventional kinematic controller such as PD ar PID and renders those controllers unsuitible for controlling the manipulator in high speed motion. Dynamic control may be required for improving the accuracy of the high speed tracking motion.
The 4th order and non-linear coupled dynamics, still remained despite of the simplified dynamic structure, of the flexible joint balanced manipulator and the uncertain kinematic and dynamic parameters may inhibit the use of mathematical inverse dynamics control law.
Considering all these aspects, this study attempted to use the artificial neural network(NN) controllers to realize the dynamic control of the flexible joint manipulators. The feedforward and feedforward-feedback types of NN controllers were proposed and tested through simulation experiments. The feedforward NN controller showed much good performances over the conventional PD controller, but the feedforward-feedback NN controller revealed much inferior performances compared to the feedforward NN controller.