A novel calibration method of 6-DOF parallel kinematic mechanisms using artificial deep neural networks

This paper proposes a novel, more efficient calibration method using an artificial deep neural network(DNN). Parallel kinematic mechanisms(PKMs) offer enhanced accuracy and repeatability due to the inherent stiffness and rigidity. However, the absolute pose error(APE) of the system scales exponentially as the size of the mechanism increases, considering geometrical and non-geometrical errors. This infers that achieving high precision necessitates meticulous calibration through iterations to mitigate error sources, especially for extra-large mechanisms. DNN models the complex relationship between input parameters and platform positions to reduce the number of pose measurements in the calibration process. This is achieved by modeling PKMs with an analytical approach to generate the training dataset for the DNN model. Taking a 6-UCU Stewart platform as an example, experimental results demonstrate the efficacy of the proposed approach on extra-large PKMs, highlighting an increase in calibration efficiency and a calibration validation of APE decrement from 3.55mm to 0.41mm.