Abstract: In the machining process,the change of cutting conditions will cause the change of data distribution,and the data under new cutting conditions is difficult to obtain.The direct use of historical data will lead to poor tool wear prediction effect.Aiming at the above problem,a small data tool wear prediction method was proposed based on domain generalization and meta-learning for tool wear prediction under varying cutting conditions.In data split stage,the historical data of cutting process were divided into pre-training set and fine-tuning set based on the cutting parameters.In the pre-training stage,the model was generalized by training sub-tasks iteratively.In the fine-tuning stage,the pre-trained model was generalized to the target data to make it better adapt to the tool wear prediction task of the target cutting condition.The proposed method was compared with multiple methods on NASA milling data set and Turbine Rotor Slot Milling data set.The results showed that the prediction effect of the proposed method were significantly better than those of the comparison methods,and it had strong universality,indicating that the proposed method could well solve the problem of tool wear prediction under varying cutting conditions.

Key words: varying cutting condition, domain generalization, meta learning, small data, tool wear prediction

摘要： 在机械加工过程中,工况变化会引起数据分布变化,且新工况下数据难以获得,直接使用历史数据会导致刀具磨损预测效果变差。针对上述问题,提出一种基于域泛化和元学习的小样本刀具磨损预测方法,对变工况下的刀具磨损进行预测。以工艺参数为依据将切削过程历史数据划分为预训练集和微调集,在预训练阶段通过迭代训练子任务的方式提高模型泛化能力,最后通过模型微调将预训练模型泛化到目标工况数据中,使其更好地适应目标工况的刀具磨损预测任务。在NASA铣削数据集和汽轮机转子轮槽铣削数据集上将所提方法与多种方法进行对比,结果显示所提方法的预测效果显著优于对比方法,具有较强的普适性,表明所提方法能很好地解决变工况背景下的刀具磨损预测问题。

关键词: 变工况, 域泛化, 元学习, 小样本, 刀具磨损预测