基于深度迁移学习的跨工况在线刀具磨损预测  

A Method for Predicting Tool Wear in Different Working Conditions Online Based on Deep Transfer Learning

作　　者：王一帆 孙江 杨恒 杨旭[1,2,3,4] 高洁 WANG Yifan;SUN Jiang;YANG Heng;YANG Xu;GAO Jie(Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110016 China;Institute of Robotics and Intelligent Manufacturing Innovation,Shenyang 110069,China;不详)

机构地区：[1]中国科学院沈阳自动化研究所,沈阳110016 [2]机器人与智能制造创新研究院,[JP]沈阳110069 [3]中国科学院大学,北京100049 [4]辽宁省智能检测与装备技术重点实验室,沈阳110179 [5]航空工业庆安集团有限公司,西安710003

出　　处：《组合机床与自动化加工技术》2025年第1期160-164,共5页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金面上项目(62073312);辽宁省基础研究计划项目(2023JH2/101300148,2023JH2/101300228,2023JH2/101300143);辽宁省自然科学基金面上项目(2022-MS-033)。

摘　　要：随着深度学习在工业中的广泛应用,通过精确可靠的刀具剩余寿命预测,可以有效地提高刀具的利用率,进而最大程度地提高加工可靠性并降低生产成本。然而,传统的刀具磨损预测通常是在相同工况下进行,致使在工况变化后预测精度较差。针对加工条件的复杂性和多样性,提出了一种用于跨工况进行刀具磨损状态预测的方法,以实现跨工况条件下精准预测刀具磨损量。针对刀具多传感器数据提出了一种提取原始数据特征的多尺度卷积神经网络,以高效、精准提取刀具数据特征,进而使多尺度卷积神经网络和Transformer训练得到具有强泛化能力的模型。最后通过共享的参数信息在新工况的下游任务中进行预测。实验中以IEEE PHM 2010挑战数据集为源数据集,以UC Berkeley Milling数据集为目标数据集进行验证。实验结果表明,在跨工况条件下能够实现精准刀具磨损预测,证实了方法的鲁棒性和可靠性。With the widespread application of deep learning in industry,accurate and reliable prediction of tool remaining life can effectively improve tool utilization,thereby greatly enhancing machining reliability and reducing production costs.However,traditional tool wear prediction is usually conducted under the same working conditions,leading to poor prediction accuracy when conditions change.To address the complexity and diversity of machining conditions,a method for predicting tool wear status across different working conditions is proposed to achieve accurate prediction of tool wear under varying conditions.A multi-scale convolutional neural network is proposed for extracting features from raw data of multiple sensors on the tool,enabling efficient and precise feature extraction.This network,along with a Transformer,is trained to develop a model with strong generalization capability.Finally,shared parameter information is used for prediction in downstream tasks under new working conditions.Experiments using the IEEE PHM 2010 challenge dataset as the source dataset and the UC Berkeley Milling dataset as the target dataset were conducted.The results demonstrate that this method can achieve precise tool wear prediction across different working conditions,confirming its robustness and reliability.

关 键 词：迁移学习 刀具 刀具磨损预测 深度学习 智能检测 

分 类 号：TH165[机械工程—机械制造及自动化] TG71[金属学及工艺—刀具与模具]