Neural network-based compensation of End position and attitude error in digital twin model of industrial robots

This paper takes the IRB-120 robot manufactured by ABB as an example and focuses on how to solve the discrepancy between the digital twin model and the end position of existing industrial robots. In this paper, a digital twin model is established using theoretical parameters, and a PSO-RBF neural network is used to compensate the error between the digital twin model and the actual robot end position and orientation. The training of the neural network takes the joint angles obtained from the inverse solution of the theoretical kinematic model as the output data, and the theoretical joint angles obtained by the twin model through the fieldbus as the input data. This implements the training process of the neural network. The twin system predicts the theoretical joint angles of the robot in real time and uses these predicted angles to drive the virtual model to ensure that its final position is consistent with that of the actual robot. Experimental validation shows that this approach reduces the error between the end position of the twin model and the end position of the actual robot from 1.4573 mm to 0.0012 mm. In addition, the orientation RPY angles were corrected from 0.0037°, 0.0091° and 0.0032° to 0.0006.