Scholarlink

<p>Tool wear is inevitable in micro-milling. The small size of micro-milling tools makes it difficult to detect wear without without using supercomputing and high-precision equipment during downtime, which seriously affects machining quality and efficiency. Therefore, accurate and real-time monitoring of micro-milling tool wear is crucial. In this paper, we propose an improved Pyramid Transformer based on global average pooling (GPVT). To address the problem of high computational complexity of traditional MHSA in processing high-speed cutting images, we design Global Average Pooling with MHSA (GAP-MHSA), significantly reducing computational overhead of model by utilizing normalization and average pooling, which satisfies stringent monitoring efficiency requirements of micro-milling. To enhance model’s ability to analyze wear characteristics of micro-milling tools, KAN linear layer is combined with MLP to make model more interpretable. In addition, considering the multi-scale characteristics of micro-milling cutter wear features, the channel convolutional attention mechanism (CCAM) is designed to integrate advantages of soft attention and hard attention to realize efficient interaction between global and local features, and ensure that the model accurately captures the wear features at different scales. The experimental results show that GPVT is more accurate than other common methods and more efficient than Transformer-based model, which meets the real-time requirements of micro-milling.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

High-accuracy tool wear monitoring for micro-milling method based on improved Pyramid Vision Transformer

  • 期刊论文

  • 发表日期：

    2025年7月15日

查看全文
  • Gang Zheng,
  • Minshu Chen,
  • Qiong Zhou

摘要

Tool wear is inevitable in micro-milling. The small size of micro-milling tools makes it difficult to detect wear without without using supercomputing and high-precision equipment during downtime, which seriously affects machining quality and efficiency. Therefore, accurate and real-time monitoring of micro-milling tool wear is crucial. In this paper, we propose an improved Pyramid Transformer based on global average pooling (GPVT). To address the problem of high computational complexity of traditional MHSA in processing high-speed cutting images, we design Global Average Pooling with MHSA (GAP-MHSA), significantly reducing computational overhead of model by utilizing normalization and average pooling, which satisfies stringent monitoring efficiency requirements of micro-milling. To enhance model’s ability to analyze wear characteristics of micro-milling tools, KAN linear layer is combined with MLP to make model more interpretable. In addition, considering the multi-scale characteristics of micro-milling cutter wear features, the channel convolutional attention mechanism (CCAM) is designed to integrate advantages of soft attention and hard attention to realize efficient interaction between global and local features, and ensure that the model accurately captures the wear features at different scales. The experimental results show that GPVT is more accurate than other common methods and more efficient than Transformer-based model, which meets the real-time requirements of micro-milling.