基于特征融合与域自适应的刀具磨损在线监测
Online Tool Wear Monitoring Based on Feature Fusion and Domain Adaptation
柳大虎 1汪永超 1何欢1
作者信息
- 1. 四川大学机械工程学院,成都 610065;宜宾四川大学产业技术研究院,宜宾 610064
- 折叠
摘要
机床状态监测对于机床健康管理以及保证工件加工质量具有重要意义.针对现有刀具磨损预测模型存在训练时间长、收敛速度慢以及泛化能力弱等问题,提出了一种分布式一维卷积神经网络对刀具磨损进行预测.采用残差连接与通道注意力模块顺序堆叠的方式作为特征提取模块,并通过交叉验证以选择合适的网络层数.由于不同传感器所提取到的特征信息可能存在冗余,使用权重差异策略以提高特征提取的有效性以及全面性.此外,考虑到训练集与测试集分布可能存在差异从而影响模型的泛化性能,引入了域自适应方法提高模型在未知数据集中的表现.为验证模型效果,使用PHM 2010 铣刀磨损数据集进行实验.实验结果表明,该模型在C1、C4、C6 三把刀具上的平均RMSE和平均MAE分别为6.97 和6.29,与TCN、TDConvLSTM等模型相比有12%以上的提升.
Abstract
Machine tool condition monitoring is of great significance for machine tool health management and workpiece processing quality assurance.In view of the problems of existing tool wear prediction models such as long training time,slow convergence speed and weak generalization ability,this paper proposes a distributed one-dimensional convolutional neural network to predict tool wear.The residual connection and channel attention modules are sequentially stacked as the feature extraction module,and cross-validation was used to select the appropriate number of network layers.Since the feature information extracted by dif-ferent sensors may be redundant,a weight difference strategy was used to improve the effectiveness and comprehensiveness of feature extraction.In addition,considering that the distribution of the training set and the test set may be different,which affects the generalization performance of the model,a domain adapta-tion method is introduced to improve the performance of the model in unknown data sets.In order to verify the effect of the model,experiments were carried out using the PHM 2010 milling cutter wear data set.The experimental results show that the average RMSE and average MAE of the model on the three tools C1,C4,and C6 are 6.97 and 6.29,respectively,which is more than 12%improved compared with TCN,TDConv-LSTM and other models.
关键词
刀具磨损监测/多传感器特征融合/权重差异策略/域自适应Key words
tool wear monitoring/multi-sensor feature fusion/weight difference strategy/domain adaptatio引用本文复制引用
基金项目
国家自然科学基金资助项目(51875370)
出版年
2024
NETL
NSTL
万方数据