It has been proved that minimizing the condition
number of the observation matrix, which is calculated from the
robot dynamic model and the associated exciting trajectories,
is very effective for improving the identification accuracy of
robotic dynamics. A relative simple dynamic model is beneficial
for reduction of the associated condition number, and hence,
several model simplification methods have been proposed in
the literature. However, the existed methods cannot be used to
efficiently process model structural errors, which will inevitably
cause inaccurate estimation of the dynamics. Therefore, a novel
model simplification method based on relative contribution of
the undetermined parameters, is proposed to overcome the
deficiency. Firstly, exciting trajectories for model simplification
are designed by using finite Fourier series and optimized by
using the condition number criteria. Then, the optimized exciting trajectory is implemented on the robot, and joint torques
and motion data are recorded, which are used to calculate
relative contribution of the undetermined parameters to joint
torques. The model can be simplified repeatedly by neglecting
the parameter that contributes least until the condition number
is small enough. Finally, the performance of the proposed
method is demonstrated by the identification and validation
experiments conducted on a lower limb rehabilitation robot.