基于多通道一维卷积神经网络的刀具磨损动态预测模型  

Dynamic prediction model for tool wear based on a multi-channel one-dimensional convolutional neural network

作　　者：黄华[1] 姚嘉靖 王永和 吕延军[2] HUANG Hua;YAO Jiajing;WANG Yonghe;Lü Yanjun(School of Mechanical Engineering,Lanzhou University of Technology,Lanzhou 730050,China;School of Mechanical and Precision Instrument Engineering,Xi'an University of Technology,Xi*an 710048,China)

机构地区：[1]兰州理工大学机电工程学院,兰州730050 [2]西安理工大学机械与精密仪器工程学院,西安710048

出　　处：《振动与冲击》2023年第2期60-67,共8页Journal of Vibration and Shock

基　　金：国家自然科学基金(51965037;51565030)。

摘　　要：针对同一工况下不同刀具磨损预测建模中的数据分布不同,从而导致的历史模型失效问题,提出了一种基于多通道一维卷积神经网络的刀具磨损动态预测建模方法。历史刀具磨损数据训练的多通道一维卷积神经网络,作为初始的刀具磨损预测历史模型。最大均值差异(maximum mean difference, MMD)法对不同刀具磨损数据进行相似度检测,当相似度相差较大时,在历史模型的基础上进行迭代更新,更新后的模型再对磨损数据进行预测。铣削试验验证结果表明,该方法能够准确预测不同刀具的磨损值大小,具有较好的自适应能力。Aiming at the failure of a historical model caused by different data distribution in the wear prediction modeling of different tool under the same working condition,a tool wear dynamic prediction modeling method based on a multi-channel one-dimensional convolutional neural network was proposed・The multi-channel one-dimensional convolutional neural network trained with historical tool wear data served as the historical model of the initial tool wear prediction.The maximum mean difference(MMD)method was used to detect the similarity of different tool wear data.When the similarity difference was large,the iterative updating was carried out on the basis of the historical model,and then the updated model was used to predict the wear data.The results of milling experiments show that the method can accurately predict the wear degree of different tools and has great adaptive ability.

关 键 词：刀具磨损 动态建模 一维卷积神经网络 最大均值差异(MMD) 

分 类 号：TH164[机械工程—机械制造及自动化]