基于融合特征与注意力网络的刀具状态监测  

Tool Condition Monitoring Based on Fusion Feature and Attention Network

作　　者：王楠 钱炜[1] 江小辉[1] 郭维诚 李星 WANG Nan;QIAN Wei;JIANG Xiaohui;GUO Weicheng;LI Xing(School of Mechanical Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Beijing Spacecraft Manufacturing Co.,Ltd.,Beijing 100094,China)

机构地区：[1]上海理工大学机械工程学院,上海200093 [2]北京卫星制造厂有限公司,北京100094

出　　处：《组合机床与自动化加工技术》2024年第10期82-88,94,共8页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金项目(52175427,52105470)。

摘　　要：刀具磨损是影响零件加工质量的重要因素之一,为了准确可靠地监测刀具磨损状态,提出了一种基于融合特征和注意力机制卷积神经网络的刀具状态预测模型。首先,在铣削过程中采取切削力与振动信号,提取了有效的加工信号;其次,利用对称点模式、小波包分解等信号处理技术对力与振动信号进行重构,建立了不同磨损状态下的信号特征二维图,并对不同工艺参数进行灰度化表征;最后,构建基于SE注意力机制的卷积神经网络模型,利用融合的信号特征图与工艺参数灰度图对刀具磨损状态进行预测。结果表明,基于融合特征与注意力网络的识别模型对于刀具的磨损状态预测有较好的识别效果。Tool wear is one of the important factors affecting the machining quality of parts.In order to accurately and reliably monitor the tool wear state,a tool state prediction model based on fusion feature and attention mechanism convolutional neural network is proposed.Firstly,cutting force and vibration signals are collected in the process of milling to extract effective machining signals.Secondly,the force and vibration signals are reconstructed by signal processing techniques,such as symmetric dot pattern and wavelet packet decomposition,to establish two-dimensional diagrams of signal characteristics in different wear States,and different process parameters are characterized by gray scale.Finally,the convolution neural network model of SE attention mechanism is constructed,and the tool wear state is predicted based on the fused signal feature map and process parameter gray map.The results show that the recognition model based on fusion features and attention network has a good recognition effect on the prediction of tool wear state.

关 键 词：刀具磨损 卷积神经网络 注意力机制 对称点模式 特征融合 

分 类 号：TH16[机械工程—机械制造及自动化] TG71[金属学及工艺—刀具与模具]