基于CEEMDAN和改进轻量化时空网络的刀具状态监测
Tool Condition Monitoring Based on CEEMDAN and BiLSTM-SN-ECA
周鹏博 1刘德平1
作者信息
- 1. 郑州大学机械与动力工程学院,郑州 450001
- 折叠
摘要
针对刀具退化特征提取困难和传统时空网络模型参数多等问题,提出了基于自适应噪声完备经验模态分解(CEEMDAN)和改进轻量化时空网络(BiLSTM-SN-ECA)的刀具磨损监测模型.首先,将刀具振动信号经CEEMADAN分解得到若干模态分量,将模态分量与振动信号结合,构造特征矩阵;其次,利用ECA改进ShuffleNetv2 基本单元,并优化ShuffleNetv2 整体结构,构造BiLSTM-SN-ECA网络模型;最后,将特征矩阵输入模型进行特征学习与磨损预测.所提方法预测值的平均绝对误差和均方根误差分别为1.246 μm和2.065 μm,结果表明该方法在减少传统时空网络模型参数量与训练时间的同时,提高了预测准确度.
Abstract
Aiming at the difficulties in extracting tool degradation features and the many parameters of tra-ditional spatiotemporal network models,a tool wear monitoring model based on adaptive noise-complete empirical mode decomposition(CEEMDAN)and improved lightweight spatiotemporal network(BiLSTM-SN-ECA)is proposed.Firstly,the tool vibration signal is decomposed by CEEMADAN,and the modal component is combined with the vibration signal to construct a feature matrix.Secondly,ECA is used to im-prove the basic unit of ShuffleNetv2,and the overall structure of ShuffleNetv2 is optimized to construct the BiLSTM-SN-ECA network model.Finally,the feature matrix is input into the model for feature learning and wear prediction.The MAE and RMSE of the predicted values of the proposed method are 1.246 μm and 2.065 μm,and the results show that the proposed method can reduce the number of parameters and training time of the traditional spatiotemporal network model,and improve the prediction accuracy.
关键词
刀具磨损监测/自适应噪声完备经验模态分解/轻量化时空网络/注意力机制Key words
tool wear monitoring/adaptive noise complete empirical mode decomposition/lightweight spa-tiotemporal network/attention mechanism引用本文复制引用
基金项目
河南省科技重大专项(171100210300-01)
出版年
2024
NETL
NSTL
万方数据