Tool wear is an inevitable phenomenon in the machining process. By monitoring the wear state of a tool, the machining system can give early warning and make advance decisions, which effectively ensures improved machining quality and production efficiency. In the past two decades, scholars have conducted extensive research on tool wear monitoring (TWM) and obtained a series of remarkable research achievements. However, physics-based models have difficulty predicting tool wear accurately. Meanwhile, the diversity of actual machining environments further limits the application of physical models. Data-driven models can establish the deep mapping relationship between signals and tool wear, but they only fit trained data well. They still have difficulty adapting to complex machining conditions. In this paper, physics-based and data-driven TWM models are first reviewed in detail, including the factors that affect tool wear, typical data-based models, and methods for extracting and selecting features. Then, tracking research hotspots, emerging physics–data fusion models are systematically summarized.

In metal-cutting technology, milling plays an important role in the product development cycle. The accurate modeling and prediction of milling forces have always been research hotspots in this field. The mechanical model based on unit-cutting force coefficients has a high prediction accuracy of cutting forces, and it is therefore widely used in the modeling of milling forces. The calibration of the cutting force coefficients can be realized by linear regression analysis of the measured average milling forces, but it needs to carry out multiple groups of variable feed slot milling experiments with full radial depth of cut, and it cannot well represent the interaction condition in peripheral milling with a non-full radial depth of cut. In peripheral milling, the tool will inevitably deform under the influence of cutting force in the direction perpendicular to the machining surface. If the force-induced deformation is ignored, the calibration of the cutting force coefficients will be out of alignment. For this non-slot milling condition where one side of the tool is mainly stressed, a cyclic calibration method for milling force coefficients considering elastic deformation along the axial contact range is proposed. Firstly, the cutting force coefficients are preliminarily calibrated by the experimental data, and, secondly, tool deformation is calculated through a preliminarily calibrated cutting force model cycle until convergence, before cutting boundaries are updated. The cutting force coefficient is then calibrated again, and it is brought back to the cantilever beam model in order to calculate the tool deformation again. The above process is repeated until the cutting force coefficient is convergent. Finally, the cutting force coefficients are obtained in order to predict and model the milling forces.

Due to the low stiffness characteristics, thin-walled parts are prone to force-induced deformation during the peripheral milling process, which severely restricts machining accuracy and efficiency. In existing studies, for curved thin-walled parts, the Finite Element Method (FEM) is usually adopted for deformation prediction. However, the traditional FEM usually requires a considerable amount of computing time, owing to the high model complexity and batch parameter evaluations. Therefore, this study proposes a method of constructing a surrogate model based on a small amount of FEM simulation data. Firstly, a peripheral milling cutting force model is established to obtain the instantaneous milling force. Secondly, a finite element model considering the material removal effect is constructed, and an iterative solution strategy is introduced to calculate the force-induced deformation. Finally, an Enhanced Latin Hypercube Sampling (ELHS) method is used to generate training samples, and the Elliptic Basis Function Neural Network (EBFNN) is selected as the surrogate model to establish a nonlinear mapping relationship between machining parameter combinations and force-induced deformation. This method enables rapid prediction of deformation at any machining position on curved thin-walled parts, reducing the computation time from hours to seconds while maintaining prediction accuracy.