Multi-stage tool wear prediction method based on multi-dimensional digital twin

Tool wear is an inevitable concern during the machining process, impacting both the tool’s remaining service life and the precision of the workpiece. Describing wear behavior comprehensively poses challenges for both data-driven and physical models. Moreover, the wear process of cutting tools entails three stages, highlighting the need for an algorithm capable of adaptively identifying and categorizing these stages accurately. We propose an adaptive multi-stage particle filtering algorithm based on a multi-dimensional digital twin to tackle these challenges. Initially, through ensemble learning, we integrate machine learning and physical models to create a wear-oriented multidimensional digital twin model. Unlike single-mechanism or data-driven models, multidimensional digital twin models offer a comprehensive analysis of research objectives, integrating the strengths of both data and mechanism models. Later, we enhance the particle filter algorithm with Statistics Process Control (SPC) to enable it to differentiate between the various stages of tool wear. Compared to conventional prediction methods like particle filtering, this approach excels at recognizing and categorizing different stages and dynamically adjusts the state space. In conclusion, the proposed model and methodology have been rigorously evaluated by applying public datasets sourced from Prognostics and Health Management (PHM). Empirical evidence from this evaluation substantiates that the introduced approach not only efficaciously discriminates between different wear phases of the tool but also quantitatively ascertains the extent of tool wear with high precision.