采用深度学习的铣刀磨损状态预测模型  

Prediction Model of Milling Cutter Wear Status Based on Deep Learning

作　　者：戴稳 张超勇[1] 孟磊磊[1] 薛燕社 肖鹏飞[1] 尹勇[2] DAI Wen;ZHANG Chaoyong;MENG Leilei;XUE Yanshe;XIAO Pengfei;YIN Yong(State Key Laboratory of Digital Manufacturing Equipment and Technology,Huazhong University of Science and Technology,Wuhan,430074;Hubei Key Laboratory of Digital Manufacturing,Wuhan University of Technology,Wuhan, 430070)

机构地区：[1]华中科技大学数字制造装备与技术国家重点实验室,武汉430074 [2]武汉理工大学湖北省数字制造重点实验室,武汉430070

出　　处：《中国机械工程》2020年第17期2071-2078,共8页China Mechanical Engineering

基　　金：国家自然科学基金资助项目(51575211,51805330,51705263);国家自然科学基金国际(地区)合作与交流项目(51861165202)。

摘　　要：为提高刀具磨损监测的预测精度与泛化性能,研究了基于深度学习的铣刀磨损状态预测,提出了基于堆叠稀疏自动编码网络与卷积神经网络的两种预测模型。堆叠稀疏自动编码网络对特征向量进行降维并将其纳入分类器来实现预测,可避免特征选择对先验知识的依赖;卷积神经网络将铣削振动数据转化为小波尺度图并输入模型完成分类,精简了传统建模流程。最后将提出的两种模型与传统神经网络模型进行比较,验证了所提模型的效率与精度。In order to improve the prediction accuracy and generalization performance of tool wear monitoring,the milling tool wear state prediction was studied based on deep learning.Two prediction models were proposed based on stacked sparse auto-encoder network and convolutional neural network.The stack sparse auto-encoder network used dimensionality reduction processing of feature vectors and incorporated them into the classifier to achieve classification prediction,avoiding the dependence on prior knowledges in feature selection.Convolutional neural networks completed the conversion of milling vibration data into wavelet scale maps as model inputs,and greatly simplified the traditional modeling processes.Finally,the two proposed models were compared with traditional neural network models to verify the efficiency and accuracy of the proposed models.

关 键 词：刀具磨损 小波变换 自动编码器 深度学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TH164[自动化与计算机技术—计算机科学与技术]