摘要
针对变分模态分解(VMD)参数选取和刀具磨损特征提取困难等问题,提出了基于松鼠觅食算法(SSA)、VMD和双向长短期记忆网络(BiLSTM)的刀具磨损状态识别模型。首先,以包络熵为适应度函数,使用SSA优化VMD的参数,利用优化后的VMD分解刀具振动信号得到4组模态分量,并进行信号重构;其次,构建BiLSTM网络模型,并把信号模态分量、原始信号和重构信号一起构成特征矩阵输入模型当中,利用BiLSTM提取信号特征;最后,通过全连接层和Softmax层对刀具磨损状态进行识别。实验结果表明,SSA能够找到VMD最优参数组合,降低信号噪声,提出的SSA-VMD-BiLSTM模型在准确率和适应性方面优于传统的LSTM模型。
Aiming at the difficulties in parameter selection of variational modal decomposition(VMD) and tool wear feature extraction, a tool wear state recognition model based on squirrel search algorithm(SSA),VMD and bi-directional long-short term memory network(BiLSTM) is proposed.Firstly, taking the envelope entropy as the fitness function, the parameters of VMD are optimized by SSA.The optimized VMD is used to decompose the tool vibration signal, obtain four groups of modal components, and reconstruct the signal;Secondly, the BiLSTM network model is constructed, and the signal modal components, the original signal and the reconstructed signal are combined to form the feature matrix input model, and the signal features are extracted by BiLSTM;Finally, the tool wear state is identified through the full connection layer and Softmax layer.The experimental results show that SSA can find the optimal parameter combination of VMD and reduce signal noise, and the proposed SSA VMD-BiLSTM model is better than the traditional LSTM model in accuracy and adaptability.
作者
刘子旭
刘德平
LIU Zi-xu;LIU De-ping(School of Mechanical and Power Engineering,Zhengzhou University,Zhengzhou 450001,China)
出处
《组合机床与自动化加工技术》
北大核心
2023年第1期119-123,共5页
Modular Machine Tool & Automatic Manufacturing Technique
基金
河南省重大科技专项(171100210300-01)。
关键词
刀具磨损预测
双向长短期记忆网络
松鼠搜索算法
变分模态分解
tool wear prediction
bidirectional long short term memory
squirrel search algorithm
variational modal decomposition
