Accurate Model Identification for a 7-DOF Cable-Driven Manipulator

Cable-driven actuation can greatly reduce the weight and inertia of manipulators, thus effectively enhancing the payload-to-weight ratio and safety. However, due to the nonlinear elastic deformation and complex friction of cables, cable-driven manipulators are difficult to model and control accurately. To solve this problem, a comprehensive model identification approach is developed to realize motor-to-joint angle conversion, kinematic calibration with pose error compensation, and dynamic identification for a 7-DOF cable-driven manipulator. First, the conversion relation from the motor angle to the joint angle is established through the numerical fitting method with a recurrent neural network to compensate for angular conversion error. Second, based on the product-of-exponentials formula, an error model with minimal kinematic parameters is formulated for kinematic calibration, and a fully connected neural network is further designed for compensating for residual pose errors caused by the influence of nongeometric factors. Third, considering nonlinear viscous friction and Stribeck effect, which are commonly observed in practical situations, a three-loop iterative scheme for dynamic identification is presented so that the mass-inertial parameters and nonlinear friction parameters can be identified simultaneously. Experiments on a self-developed 7-DOF cable-driven manipulator demonstrated that the proposed model identification approach could be used to accurately acquire kinematic and dynamic models that better accord with the actual situation.