关键词: 刀具剩余寿命预测, 加工过程监控, 迁移学习, 长短时记忆网络, 动态对抗域适应, 轮槽铣刀

Abstract: Tool Remaining Useful Life (RUL) prediction model of historical processing conditions tends to fail to predict tool RUL under new processing conditions,and there are not sufficient training data to build a new prediction model.Thus,a transfer learning method based on dynamic adversarial domain adaptation was proposed to quickly build a tool RUL prediction model under new processing conditions.The tool RUL prediction model in source domain was pre-trained by process monitoring data sample with RUL labels under historical processing conditions.Adversarial domain adaptation was implemented using a small number of target domain samples under new process conditions,parts of the pre-trained model parameters are adjusted.The adjusted model was used to predict the tool RUL under the new processing conditions.The target domain samples were updated,and the adversarial domain adaptation and RUL prediction were repeated until the end.The effectiveness of proposed method was verified by an illustrative example of slotting cutter.

Key words: tool remaining useful life prediction, process monitoring, transfer learning, long short-term memory network, dynamic adversarial domain adaptation, slotting cutter