Research on tool wear state identification method driven by multi-source information fusion and multi-dimension attention mechanism

Peining Wei,Rongyi Li,Xianli Liu,Haining Gao,Mingqiu Dai,Yuhan Zhang,Wenkai Zhao,Erliang Liu

DOI: https://doi.org/10.1016/j.rcim.2024.102741

IF: 10.103

2024-02-26

Robotics and Computer-Integrated Manufacturing

Abstract: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.

robotics,computer science, interdisciplinary applications,engineering, manufacturing

What problem does this paper attempt to address?

The paper aims to address several key issues in the monitoring of cutting tool wear states:

1. **Information Redundancy and Lack**: The redundancy and lack of information when extracting cutting tool wear features lead to inaccurate estimation of tool wear states.
2. **Transition Phase Fuzzy Area**: The fuzzy area during the transition phase of tool wear is a key factor causing misjudgment of tool wear states.

To solve these problems, the study proposes a data-driven intelligent tool wear state detection technology based on multi-source information fusion and multi-dimensional attention mechanisms. Specifically, this method improves the recognition of tool wear states through the following ways:

- **Multi-source Data Fusion**: Utilizing data from multiple sensors to achieve information complementation, which enhances the reliability and robustness of the system.
- **Channel-Spatial Attention Mechanism**: Introducing a channel-spatial attention mechanism to selectively learn various features related to cutting tool wear, reducing the impact of fuzzy areas between different wear stages.

Experimental results show that the proposed method can accurately identify tool wear states based on process data and allows for more flexible and precise cutting tool replacement decisions during the machining process. The comprehensive recognition accuracy of the model is 0.982, with F1 scores of 0.977, 0.968, and 0.993 in the initial, normal, and severe wear stages, respectively, outperforming other models.