A Semilinearized Approach for Dynamic Identification of Manipulator Based on Nonlinear Friction Model

Abstract

Accurate identification of manipulator dynamics parameters is crucial for achieving precise control and optimal performance in various robotic applications. Existing methods primarily utilize least squares or weighted least squares for dynamic parameters identification which cannot effectively integrate nonlinear friction model and satisfy physical feasibility constraints. Consequently, this study proposed an iteratively semi-linearized approach for identifying dynamic parameters of manipulator based on nonlinear friction models. First, the nonlinear friction model parameters of joint are pre-identified to prepare for feasibility base parameter estimation. Secondly, to establish a semi-linearized dynamic model, the robot and friction torque are separated, allowing for the linearization of the robot dynamic model while preserving the nonlinearity of the friction model. Utilizing this semi-linearized model, an iterative identification network is established to accurately identify dynamic parameters. This method effectively satisfies physical feasibility constraints and integrates various nonlinear friction models, resulting in improved identification accuracy. Furthermore, it seamlessly incorporates geometric metrics to better align the identified dynamic parameters with the actual physical system. Finally, the utilization of neural network-based nonlinear compensation technology further reduces identification errors. A series of experiments were conducted using a self-developed nine-degrees-of-freedom redundant manipulator. The experimental results demonstrated the efficiency and superiority of the proposed algorithm through performance comparison with various methods and friction models. A comprehensive analysis and discussion have revealed the underlying principles of this approach.