特征融合与BP神经网络结合的刀具磨损预测  

Tool Wear Prediction Based on Feature Fusion and BP Neural Network

作　　者：郭宏[1] 徐延 伊亚聪 胡孔耀 GUO Hong;XU Yan;YI Yacong;HU Kongyao(School of Mechanical Engineering,Taiyuan University of Science and Technology,Shanxi Taiyuan 030024,China)

机构地区：[1]太原科技大学机械工程学院,山西太原030024

出　　处：《机械设计与制造》2025年第1期108-111,116,共5页Machinery Design & Manufacture

基　　金：山西省回国留学人员科研教研资助项目(HGKY2019079)。

摘　　要：通过监测刀具磨损情况,能够有效应对生产加工中的意外状况。为了精确监测刀具的磨损状态,提出了一种多传感器特征融合及BP神经网络结合的刀具磨损预测方法。首先对工业加工中采集到的切削力、振动、声发射信号进行小波阈值去噪,然后在时域、频域和时频域内分析并提取特征,再将融合后的各类传感器特征使用Pearson相关系数和主成分分析(PCA)实现数据降维,最后将降维后的融合特征输入搭建好的BP神经网络,通过非线性仿真分析,从而实现刀具磨损量的预测。案例验证表明:与单一传感器预测相比,提出的多传感器特征融合的刀具磨损预测方法误差最小,且决定系数R2达到0.993。By monitoring tool wear,it can effectively respond to unexpected conditions in production and processing.In order to accurately monitor the wear status of the tool,a tool wear prediction method based on multi-sensor feature fusion and BP neural network is proposed.First,perform wavelet threshold denoising on the cutting force,vibration,and acoustic emission signals col⁃lected in industrial processing,then analyze and extract features in the time domain,frequency domain,and time-frequency do⁃main,and then use Pearson correlation for the fusion of various sensor features Coefficients and principal component analysis(PCA)realize data dimensionality reduction,and finally input the fusion features after dimensionality reduction into a built BP neural network,and use nonlinear simulation analysis to predict the amount of tool wear.Case verification shows that:compared with single sensor prediction,the proposed multi-sensor feature fusion tool wear prediction method has the smallest error,and the coefficient of determination R2 reaches 0.993.

关 键 词：传感器 特征提取 小波去噪 PCA BP神经网络 磨损预测 

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]