Abstract: Aiming at the problems of insufficient extraction of tool health status features and low recognition accuracy of CNC machine tools, this paper proposes a tool health status monitoring method. Firstly, the improved complete set empirical mode decomposition (ICEEMDAN) and the fine composite multi-scale weighted permutation entropy (RCMWPE) are used for signal processing and feature extraction, and the feature information that can characterize the health state of the tool is extracted. Secondly, by t-SNE processing, the dimensionality reduction and fusion of feature information are realized. Finally, the obtained low-dimensional features are input into the constructed whale-optimized kernel extreme learning machine (WOA-KELM) health recognition model, so as to classify and identify the tool health state. Experimental verification shows that the proposed signal processing, feature extraction and state recognition model have achieved good results in tool health monitoring, and its state recognition accuracy is as high as 99.76%, which can efficiently and accurately classify and identify tool wear status.