Toward digital twins for high-performance manufacturing: Tool wear monitoring in high-speed milling of thin-walled parts using domain knowledge

The condition of cutting tool in the high-performance machining of aerospace Ti6Al4V thin-walled parts is a key factor in determining service performance and productivity. However, the complex dynamic characteristics of thin-walled cutting systems lead to a different evolution of the machining signals than in conventional machining, which drastically reduces the performance of the extracted features. This results in the faint features that are exclusive to the machining of thin-walled components. To establish a data-driven mapping between the faint signal features and tool wear conditions in thin-walled parts cutting, this study proposes a domain-knowledge based-tool wear monitoring method, which efficiently addresses the effects of random perturbations caused by the weak stiffness of workpieces. Firstly, to eliminate the influence of mutated vibration, a quantitative characterization model for the evolution of cutting data is proposed by non-linearly superimposing multi-dimensional behavioral indicators of non-normally distributed feature vectors. Secondly, the wear law of milling tools is determined from the wear mechanism on the rake and flank faces, which avoids the effect of human factors on the measurement of data labels. Based on this, the lightweight tool wear recognition model is then developed through the filtered features with domain knowledge. According to the results, the performance of cutting force in the x-direction and the axial bending moment are more than 6 % ahead of other channels. Time domain features and time-frequency domain features are advantageous in monitoring tool status through regression and classification, respectively. Experiments demonstrate that the proposed method improves the accuracy by 7 % while reducing the number of model parameters by more than 5.65 times. This may provide an efficient way to build digital twins for high-performance machining. Moreover, a dataset containing the machining data of multiple milling tools is provided, which lays the foundation for forward research on tool condition monitoring and anomaly detection in high-performance cutting.