16 June 2023 in IEEE/ASME Transactions on Mechatronics
An elaborated bimanual 4-degrees-of-freedom (DOF) parallel robot is designed with large translational workspace, light moving mass and direct-drive actuation. The transient response greatly affects the throughput when using this robot for high-speed pick-and-place applications. However, this cannot be properly taken care by existing black-box model-based control schemes. Meanwhile, it is tedious to derive its dynamic model analytically and such a model is inapplicable for the real-time control. In this work, a gray-box-model-based control structure is proposed, with the retained inertial dynamics are directly derived by the principle of virtual work and the other parts are estimated by adaptive neural networks. This ensures calculation efficiency and integrity of the dynamics. Moreover, a prescribed performance function is constructed to ensure the specified tracking requirements both in transient and steady states. On the basis, the robust integral of signum of error term is incorporated to compensate the structural and unstructural uncertainties, which further improves the robustness during high-frequency motion. Comparative real-time experiments have been performed on the actual robot with attainment of the predefined performance and higher tracking accuracy.
4 March 2022 in IEEE Transactions on Instrumentation and Measurement
The newly developed 4PPa-2PaR parallel mechanism has subchains on its limbs, and the imperfect assembly process causes the kinematic error. The conventional matrix-based method is difficult to set up its kinematic error model, as the kinematic constraints cannot be represented by the homogenous transformation matrix intuitively. In this work, the actual kinematic model is established using a vector-based method, considering the nonideal lengths of links and misalignment of active joints’ home sensors. Thereby, the kinematic constraints of the mechanism are intuitively incorporated into the model, and the deviations of kinematic parameters are identified by the Levenberg–Marquardt algorithm. To achieve error compensation using the inverse of the actual kinematic model, a flange with three collinear reflector holders is designed and mounted on the moving platform, and a three-point measurement method is utilized to capture its pose without using a tracker-machine control sensor (T-MAC). Another flange for reflector holders is designed and mounted on the active joint for setting up the base frame so that all measurement data are transformed from the laser tracker’s frame to the base frame to complete the calibration process. Experiments are conducted on the testbed to validate the effectiveness of the proposed calibration method.