Kinematic and Nonparametric Calibration of a Parallel Machining Robot Based on an Artificial Neural Network

Abstract

Model-based kinematic calibration is feasible to improve robot accuracy by compensating for the geometric errors. However, non-geometric errors such as backlash and deformations cannot be eliminated via this approach since they are difficult or impossible to fully model. This paper proposes a two-stage calibration method to comprehensively compensate for the geometric and non-geometric error sources for a 5-DoF parallel machining robot. In the first stage, the geometric error propagation model of the parallel robot is established using the vector loop method. On this basis, a kinematic calibration algorithm to compensate for geometric errors is presented by employing the truncated singular value decomposition technique (TSVD) from an ill-posed identification matrix problem. To further improve the accuracy of the robot, in the second stage, an artificial neural network (ANN) is designed and applied to eliminate the residual non-geometric errors by integrating it into a kinematic calibration model. As a result, an augmented pose error model with computational efficiency is developed to accurately describe the comprehensive error characteristics of the parallel robot. Finally, simulation experiments are conducted. The results demonstrate that the presented approach is effective and robust, with average position and orientation errors reduced by 68.6% and 93.5%, respectively, compared with model-based kinematic calibration.