Abstract:

To address the issues of high workload and low accuracy in existing machine vision-based off-line tool wear measurement， an in-situ monitoring method for milling tool wear in machining centers was proposed based on a spatial attention mechanism U-Net. Firstly， an automatic tool wear monitoring experimental platform was established. Through PMC programming， the platform employed NC codes to control spindle orientation and rotation angles， enabling automatic positioning of the wear areas on the lateral cutting edges of the milling cutters. Secondly， communication with the machine tool through the Focas protocol and C# scripts， automatic in-situ imaging of the milling cutter's bottom edge and each lateral edge were realized. Next， label files were created by using LabelMe software， the spatial attention mechanism U-Net semantic segmentation method was employed to accurately identify the wear areas， and the quantifiable tool wear values were obtained by combining the morphological method. Finally， the proposed model was compared with semantic segmentation models such as Deeplabv3+， full convolutional networks（FCN）， Lraspp， SegNet， and PspNet to verify the effectiveness and accuracy of the proposed method.

Key words: U-Net model, attention mechanism, machine vision, milling tool, wear monitoring

摘要：

针对目前基于机器视觉刀具磨损离线测量工作量大、准确度差的问题，提出一种基于空间注意力机制U-Net的加工中心铣刀磨损在位监测方法。首先，搭建了刀具磨损量自动监测实验平台，该平台通过PMC编程，使用NC代码控制主轴定向和旋转角度，实现对铣刀侧刃磨损区域的自动定位；其次，通过Focas协议和C#脚本与机床通信，实现铣刀底刃和每个侧刃的自动在位拍摄；然后，通过LabelMe软件制作标签文件，采用空间注意力机制U-Net语义分割方法精确识别磨损区域，结合形态学方法计算获得量化的刀具磨损值；最后，将提出模型与Deeplabv3+、全卷积网络（FCN）、Lraspp、SegNet、PspNet等语义分割模型对比，验证所提方法的有效性和准确性。

关键词: U-Net模型, 注意力机制, 机器视觉, 铣刀, 磨损监测