基于人工智能的切削刀具疲劳强度预测方法  

Prediction method of cutting tool fatigue strength based on artificial intelligence

作　　者：王荔檬 赵莹莹 杨铨[1] WANG Limeng;ZHAO Yingying;YANG Quan(Guangxi Vocational&Technical Institute of Industry,Nanning 530001,Guangxi,China)

机构地区：[1]广西工业职业技术学院,广西南宁530001

出　　处：《模具技术》2024年第6期65-72,共8页Die and Mould Technology

基　　金：广西重点研发计划机械门锁复合防盗锁芯结构及其防盗方法的研究与应用(编号:2023AB01165);广西重点研发计划智慧农机动力域控制系统关键技术开发及应用(编号:2021AB01008)。

摘　　要：研究基于人工智能的切削刀具疲劳强度预测。采用小波降噪的方法过滤信号噪声并去除包含的异常值;以时域和频域特征为目标,提取更加完整的信号特征,利用Pearson系数法和MIC系数法对特征信号排序,通过核主成分分析完成信号特征融合,运用广义回归神经网络训练不同特征信号,根据获得的刀具磨损量实现对切削刀具疲劳强度的预测。实验结果表明:传统方法利用反向传播神经网络(back propagation neural network,BPNN)获得的磨损量精确度较低,预测刀具的裂纹宽度,从第45 h开始迅速扩大,工作时间超过60 h后刀头会折断;本方法在初始阶段就处理了首尾噪声和中段异常值,广义回归神经网络(generalized regression neural network,GRNN)获得的磨损量更加精准,预测刀具裂纹宽度从第75 h开始扩大,工作时间超过80 h后,刀头才会出现会折断现象,这一预测与实际结果之间的差异极小,可见本方法的预测更加精准。Traditional methods have poor ability to handle raw signal noise and outliers.Poor quality signals affect the integrity of feature extraction results,affect the prediction of tool wear,and cause the prediction of cutting tool fatigue strength to be inconsistent with the actual situation.Research is being conducted on an artificial intelligence based method for predicting cutting tool fatigue strength.Using wavelet denoising method to filter signal noise and remove contained outliers;Targeting time-domain and frequency-domain features,extract more complete signal features,use Pearson coefficient method and MIC coefficient method to sort the feature signals,complete signal feature fusion through kernel principal component analysis,train different feature signals using generalized regression neural network,and predict the fatigue strength of cutting tools based on the obtained tool wear amount.The experimental results show that the traditional method of using BPNN to obtain wear accuracy is low,and the predicted crack width of the tool rapidly expands from the 45th hour onwards.After working for more than 60 hours,the tool head will break;The research method processed the head and tail noise and mid section outliers in the initial stage,and the wear amount obtained by GRNN was more accurate.It predicted that the crack width of the tool would expand from the 75th hour,and the tool head would only break after working for more than 80 hours.The difference between this prediction and the actual results was minimal,indicating that the prediction of the method in this paper is more accurate.

关 键 词：人工智能 切削刀具 疲劳强度 预测 

分 类 号：TH181[机械工程—机械制造及自动化]