Co-Kriging model-based multi-fidelity regression for refining milling stability predictions of variable tool overhang length using limited experimental data

Determining chatter-free machining parameters is critical in process planning since chatter vibrations significantly affect machining quality and production efficiency. Stability lobe diagram is essential in selecting chatter-free machining parameters, but its predictions confront challenges in both theoretical accuracy and experimental efficiency. To address this, the Co-Kriging modelling is introduced to predict the tool overhang length-dependent stability limits with limited experiments. First, tool tip dynamics and cutting force coefficients of a source overhang length are roughly identified to obtain sufficient low-fidelity (LF) theoretical stability limits. The entropy-based sampling and affine transformation are used to sequentially obtain high-fidelity (HF) experimental stability limits and update LF stability limits. HF and updated LF datasets are combined to train a multi-fidelity (MF) Co-Kriging model. For a target overhang length, MF stability limits of the source overhang length serve as LF data, and a minimum mean absolute percentage error (MAPE)-based sampling guides the acquisition of HF stability limits. Combining HF data and LF data processed by affine transformation to train a target Co-Kriging model. Case studies reveal that the proposed method exhibits lower MAPE values and significantly outperforms other comparison methods in predicting experimental stability limits, particularly when confronted with limited experimental data.