Ta-12W tantalum-tungsten alloy is widely used in aerospace and nuclear energy fields due to its excellent high-temperature strength and corrosion resistance. However, its poor machinability often leads to severe tool wear, and the research on tool wear monitoring during its milling process is still incomplete. To address this problem, firstly, a multi-sensor acquisition system was established, and a four-blade integrated end mill was used to conduct milling experiments to collect multi-source signal data. Then, a tool wear model based on physical information was established to improve the physical interpretability and classification accuracy of the wear state. Secondly, maximum overlap discrete wavelet transform (MODWT) and random forest (RF) were used to extract features highly correlated with the wear state. Finally, a hidden Markov model based on an asynchronous learning factor particle swarm optimization algorithm (AsyLnCPSO-HMM) was proposed, which effectively avoided local optimality and achieved robust state recognition. Experimental results show that the wear state classification accuracy of the model reaches 96.11%, higher than traditional HMM and classic classification models, showing good performance and application potential.
Tang et al. (2026) studied this question.