Data-physics collaborativige fusion prediction method for tool remaining useful life based on Mamba state space and physical description

Abstract

Tool wear is an inherent phenomenon of the metal cutting process, the traditional replacement strategy relies on experience or a fixed cycle easily leads to waste of resources or workpiece damage, accurate prediction of the tool's remaining useful life (RUL) has become a key issue in the field of intelligent manufacturing urgently need to break through. Aiming at the problems of insufficient nonlinear processing capability of physical models and weak interpretability of data-driven models in the existing RUL prediction, this study proposes a data-physics collaborative fusion prediction method for tool remaining useful life based on Mamba state space and physical description. The method breaks through the traditional single-model paradigm and achieves in-depth characterization of the cutting process through a dual modeling mechanism: firstly, a time-series feature extraction network based on the Mamba state space is constructed, and a selective memory mechanism is adopted to achieve the screening of degradation features and non-linear characterization; secondly, a two-stage piecewise physical degradation model is established. The explicit mathematical expressions are deduced based on the geometrical features of the tool wear curve, and the prior distributions of the model parameters are estimated from historical data. The Particle Filter (PF) algorithm is introduced to establish a collaborative optimization mechanism for the dual models, and the physical parameters are dynamically updated through importance sampling to achieve tool RUL prediction under Data-physics collaborative fusion (DPCF). The experimental results show that the method can achieve accurate monitoring of tool RUL and has a certain reference value for efficient tool change in the metal-cutting process.