Cutting tool wear condition monitoring technology is the key technology of advanced manufacturing system and a crucial component of machining. The stage of the tool wear has a direct impact on performance of the workpiece and efficacy of machine tool. However, the duplication and lack of wear information and the fuzzy area of the tool wear transition stage are the key factors contributing to the incorrect estimation of the tool wear state when extracting cutting tool wear feature information. As a consequence, this study presents a data fusion from several sources-based intelligent tool wear state detection technique. The fusion of data from several sources can effectively realize the complementarity of machining information. This provides the model with more comprehensive identification data. The mapping between wear state and wear characteristics is precisely established. To address these issues, the attention mechanism of channel and spatial latitude is integrated into the feature extraction. The model that was constructed in the present investigation has a comprehensive identification accuracy of 0.982. The F1 score of initial, normal and severe wear stage of tool wear are 0.977, 0.968 and 0.993, which are better than other models. Experiments show that the identification method proposed in this study may provide accurate tool wear condition identification based on machining process data, allowing for more flexible and precise cutting tool change decisions in machining.

中文翻译：

---

多源信息融合和多维度注意力机制驱动的刀具磨损状态识别方法研究

刀具磨损状态监测技术是先进制造系统的关键技术，是机械加工的重要组成部分。刀具磨损的阶段直接影响工件的性能和机床的功效。然而，磨损信息的重复和缺失以及刀具磨损过渡阶段的模糊区域是在提取刀具磨损特征信息时导致刀具磨损状态错误估计的关键因素。因此，本研究提出了一种基于多个来源的数据融合的智能刀具磨损状态检测技术。多源数据的融合可以有效实现加工信息的互补。这为模型提供了更全面的识别数据。精确建立磨损状态与磨损特性之间的映射。为了解决这些问题，将通道和空间纬度的注意力机制集成到特征提取中。本次调查构建的模型综合识别精度为0.982。刀具磨损初始、正常和严重磨损阶段的F1分数分别为0.977、0.968和0.993，优于其他模型。实验表明，本研究提出的识别方法可以根据加工过程数据提供准确的刀具磨损状况识别，从而在加工过程中做出更灵活、更精确的换刀决策。