Abstract: Tool condition monitoring is a key technology for high-quality, high-efficiency, and intelligent machining. However, unsupervised monitoring remains challenging. The main reason is that unlabeled time-series signals generated during cutting are difficult to annotate accurately. An unsupervised tool condition monitoring method based on deep feature fusion and temporal clustering was proposed. The time interval between each sample and the cluster center was used to characterize temporal dependence. A time dependent penalty factor and a time-span balance coefficient were designed to strengthen the sequential characteristics of clustering results. Based on this, a time-dependent enhanced fuzzy C-means clustering algorithm was developed. An unsupervised tool condition monitoring framework was then established. High-temperature alloy milling experiments were conducted to verify the effectiveness and generalization capability of the proposed method. Compared with existing clustering methods and classification algorithms, the proposed method improves the accuracy of tool wear state self-labeling and wear state prediction by 24.55% and 15.33%, respectively. It also shows good adaptability under different cutting conditions. These results provide an effective solution for unsupervised tool wear state monitoring in cutting processes.