| 一种不同工艺条件下刀具磨损状态多类域适应迁移辨识方法 |
| |
| 引用本文： | 史珂铭,邹益胜,刘永志,丁昆,丁国富. 一种不同工艺条件下刀具磨损状态多类域适应迁移辨识方法[J]. 中国机械工程, 2022, 33(15): 1841-1849. DOI: 10.3969/j.issn.1004-132X.2022.15.010 |
| |
| 作者姓名： | 史珂铭 邹益胜 刘永志 丁昆 丁国富 |
| |
| 作者单位： | 1.西南交通大学机械工程学院，成都，6100312.西南交通大学计算机与人工智能学院，成都，610031 |
| |
| 基金项目： | 国家重点研发计划（2020YFB1708001）；四川省智能制造与机器人重大科技专项（2019ZDZX0021） |
| |
| 摘 要： | 在新的工艺条件下，针对采用历史工艺条件进行训练的刀具磨损状态辨识模型识别准确率低的问题，提出了一种基于迁移学习的跨工艺条件刀具磨损状态辨识模型。构建卷积神经网络提取刀具样本可迁移特征，利用最大均值差异测量不同工艺条件下刀具样本分布差异，通过类间-类内距离约束提升源域特征的样本距离，对目标域数据概率矩阵采取最大化核范数的策略，以提取区分性高的目标域样本故障特征。以铣刀加工试验为例验证了模型的有效性，模型的平均辨识准确率为96.8%，比没有类间-类内距离约束与最大化核范数的方法平均辨识准确率提升4.9%。
|
| 关 键 词： | 刀具磨损 工艺条件 迁移状态辨识 类间-类内距离约束 最大化核范数 |
A Multi Class Domain Adaptive Transfer Identification Method for Tool Wear States under Different Processing Conditions |
| |
| SHI Keming,ZOU Yisheng,LIU Yongzhi,DING Kun,DING Guofu. A Multi Class Domain Adaptive Transfer Identification Method for Tool Wear States under Different Processing Conditions[J]. China Mechanical Engineering, 2022, 33(15): 1841-1849. DOI: 10.3969/j.issn.1004-132X.2022.15.010 |
| |
| Authors: | SHI Keming ZOU Yisheng LIU Yongzhi DING Kun DING Guofu |
| |
| Affiliation: | 1.School of Mechanical Engineering，Southwest Jiaotong University，Chengdu，6100312.School of Computing and Artificial Intelligence，Southwest Jiaotong University，Chengdu，610031 |
| |
| Abstract: | Under the new processing conditions， aiming at the problems of low identification accuracy rate of tool wear identification model trained under historical processing conditions， a tool wear state identification model across processing conditions was proposed based on transfer learning. Firstly， convolutional neural network was constructed to extract the transfer features of tool samples， and the difference of tool sample distributions was measured by the maximum mean difference under different processing conditions. Secondly， the sample distance of source domain features was improved by IDC. The strategy of maximizing the norm was adopted for the probability matrix of target data to extract the fault features of target domain samples with high discrimination. Finally， the milling cutter machining tests were taken as an example to verify the validity of the model. The average identification accuracy rate of the model is as 96.8%， which is as 4.9% higher than that of the method without IDC and maximum kernel norm. |
| |
| Keywords: | tool wear processing condition transfer state identification inter-class-intra-class distance constraint（IDC） maximizing kernel norm |
|
| 点击此处可从《中国机械工程》浏览原始摘要信息 |
|
点击此处可从《中国机械工程》下载免费的PDF全文 |
