Abstract: Parallel robot kinematic singularities often lead to excessive motor driving torque and even damage. To reduce the failure rate of drive motors, this paper takes a 6-PSU type parallel robot as the research object and proposes a solution based on dynamic modeling and parameter identification, which can accurately estimate the driving torque in advance according to the preset trajectory for safety verification. First, combining the Kane’s method and the principle of virtual work, an inverse dynamic refined model that includes the friction of ball screw modules, servo motor friction, and inertia is established; QR decomposition is used to obtain the minimum identifiable parameter set, and based on this, considering the symmetry of the cross-section of the linkage, the negligible linkage inertia is discovered, resulting in a simplified linearized model; statistical methods are used to verify the necessity of considering friction when optimizing the excitation trajectory; a physical simulation model of the robot is built using Simulink/Multibody for parameter identification and driving torque estimation simulation experiments, and the parameter identification accuracy, driving torque estimation accuracy, and noise resistance of both the minimum identifiable parameter set model and the simplified model are quantitatively analyzed; finally, experiments are conducted on the 6-PSU type parallel robot platform, and the results show that the simplified model proposed in this paper effectively simplifies the model complexity while ensuring the same driving torque estimation accuracy, and has clear engineering value.