Sequential Physically Consistent Dynamic Parameter Identification in Parallel Robots

Abstract

Parallel robots have been widely applied because of their high stiffness and dynamic responsiveness. Advanced model-based control of such systems requires the accurate identification of dynamic parameters. This paper proposes a physically consistent dynamic parameter identification method enforced by a sequential identification procedure. Observable dynamic parameters in each actuator are first identified independently in sequence. These results are then incorporated as prior knowledge, represented as Gaussian distributions. A criterion considering sequential identification performance is also proposed to generate excitation trajectories. The experiments are conducted on a 3-degree-of-freedoms parallel robot of type 2RRU-1RRS. According to the experimental tests, the proposed method improves the accuracy of torque estimation even under insufficient excitation. It validates the effectiveness and robustness of the proposed method.