DT-CEPA: A digital twin-driven contour error prediction approach for machine tools based on hybrid modeling and sparse time series

Abstract

Contour error prediction is of great significance for optimizing machining accuracy of machine tools (MTs). However, few commercial computer numerical control (CNC) systems provide continuous interpolation data which is necessary for traditional prediction methods. Instead, only sparse time series data can be collected through the un-realtime interconnection interface of CNC systems, which makes it difficult for accurate prediction. Hence, a digital twin-driven contour error prediction approach (DT-CEPA) is proposed in this study based on hybrid modeling and sparse time series. Firstly, a DT model of each feed axis in MT is constructed to describe its dynamic characteristics based on hybrid modeling method which combines parametric model (PM) and data-based model. Particularly, to adapt to the sparse time series, the eXtreme Gradient Boosting (XGBoost) which is an ensemble learning model is employed in hybrid modeling to compensate the residual error of PM. Meanwhile, a dataset with multi-source features is constructed to improve model accuracy and stability. Therefore, the actual position of feed axis can be predicted accurately with the DT model and the corresponding contour error of actual tool path can be then calculated. Secondly, a practical updating strategy for the DT model integrating incremental learning is proposed to maintain consistency between digital space and physical space, where the updating can be adaptively triggered and implemented with an appropriate scale to ensure the model accuracy while reducing updating cost. Finally, experiments and comparative analysis are preformed to validate the performance of DT-CEPA in terms of prediction accuracy and stability.