Modeling and prediction of full-term thermal error in linear axis of machine tools based on MSTGCN-A

According to ISO 230-3, the linear axis includes six thermal error motions. Measuring and modeling multi-position full-term thermal errors in linear axis are challenging and essential since they are position-dependent and time-varying. This study presents for the first time a method for modeling and predicting the full-term thermal error of linear axis with multi-positions. Firstly, the full-term thermal error of the linear axis was measured in real-time. The nonlinearity and time delay of the thermal deformation of the screw are analyzed based on the heat transfer theory. Then, a multidimensional spatio-temporal graph convolution-attention mechanism (MSTGCN-A) model is proposed to model the full-term thermal error at multi-positions on the linear axis. The model adaptively learns the adjacency relationship of nodes and fuses the spatial information of temperature points without screening temperature-sensitive points. GRU-CNN, MTCN-A, and BiLSTM-CNN are also used as comparative models to verify the accuracy of the models. Under varying operating conditions, the proposed model outperforms the comparative model that requires temperature-sensitive point selection.