Abstract: The accuracy of the digital twin model of cutting tool established under historical working conditions is reduced,and there is a lack of sufficient training samples to retrain the model under new working conditions.For this reason,the deep transfer learning strategy was introduced into the digital twin model of cutting tool to establish the digital twin model of cutting tool under variable working conditions.Based on the edge distribution adaptation rules,the similar source domain data features and target domain data features were learned.The model trained in the source domain could be migrated to the target domain with only a few target domain samples,thereby improving the monitoring accuracy and adaptability of the digital twin model of cutting tool in the target domain.Through the experimental verification,compared with the model without migration strategy,the digital twin model with migration strategy established in this paper contributed to improve the tool wear monitoring accuracy and the adaptability under variable working conditions.

Key words: tool wear monitoring, digital twin, variable working condition, deep transfer learning, particle filter

摘要： 针对历史工况条件下建立的刀具数字孪生模型对新工况监测的准确度降低,且新工况下缺乏足够训练样本重新训练模型的问题,将深度迁移学习策略引入刀具数字孪生模型,建立刀具切削变工况数字孪生模型。该模型基于边缘分布适配规则,使模型学习到分布相似的源域数据特征和目标域数据特征,只需较少的目标域样本即可将源域训练好的模型迁移到目标域进行监测,从而提高刀具数字孪生模型在目标域的监测精度和适应性。实验验证,相比未引入迁移策略的模型,所建模型能够提高变工况条件下的磨损监测精度和对新工况的泛化能力。

关键词: 刀具磨损监测, 数字孪生, 变工况, 深度迁移学习, 粒子滤波