Multiple operational mode prediction at milling tool-tip based on transfer learning

Understanding the tool-tip dynamics is crucial for evaluating the performance in milling and essential for chatter prediction; obtaining and predicting tool-tip modes efficiently and accurately is thus essential, especially when the milling parameters or tool-holder assembly change. However, there is currently no such efficient and explainable method with high generalization ability for obtaining and predicting the tool-tip modes considering the above change. To address this issue, the stochastic subspace identification (SSI) method is initially used to acquire multiple operational modes more efficient and cost-effective than traditional methods under varying milling parameters. Subsequently, machine learning (ML) models are trained to predict the above modes under varying spindle speeds and axial cutting depth. Moreover, when changes occur in the tool-holder assembly, a transfer learning (TL) model based on receptance coupling substructure analysis (RCSA) theory is proposed to re-establish the modes prediction model efficiently with the above data. The TL model has a modal frequency prediction error below 2% and a damping ratio prediction error below 10%, thereby demonstrating robust generalization capabilities. Finally, predicting milling stability with the above modes prediction model, which can provide a stability lobe diagram with higher accuracy than the traditional method, is introduced. In conclusion, the multiple operational modes are acquired more efficiently with the SSI method, and the ML model or TL model with RCSA theory is thus established efficiently when milling parameters or tool-holder assembly change. The obtained model is used for chatter prediction as follows and performs better in prediction accuracy.