摘要
针对刀具磨损在线监测过程中信号特征较弱、外界噪声干扰较大导致的预测准确度较差问题,提出了一种基于贝叶斯优化(BO)双向长短时记忆网络(Bi-LSTM)的刀具磨损状态监测模型。结合斯皮尔曼相关系数和最大互相关系数来筛选降噪后切削力信号的时域、频域及时频域特征,输入到建立好的Bi-LSTM模型进行训练;针对Bi-LSTM模型参数组合对精度影响大且难以选择的问题,采用贝叶斯优化算法进行超参数寻优;利用铣削加工实验对模型进行验证。结果表明,该方法能快速得到模型最优超参数,同时兼具稳定性和准确性,与其他深度学习模型相比,准确率更高,实验证明了该模型的有效性和可行性。
Aiming at the problem of poor prediction accuracy caused by weak signal characteristics and large external noise interference in the process of tool wear on-line detection,a tool wear detection model based on bayesian optimized bidirectional long-term and short-term memory network is proposed.The spearman correlation coefficient and the maximum correlation number are combined to screen the time domain,frequency domain and time domain characteristics of the denoised cutting force signal,which are input into the established Bi-LSTM model for training.Aiming at the problem,the parameter combination of Bi-LSTM model has a great impact on the accuracy and is difficult to select,bayesian optimization algorithm is used to optimize the super parameters.The milling experiment is used to verify the model.The results show that this method can quickly obtain the optimal super parameters of the model,and has both stability and accuracy.Compared with other deep learning models,the accuracy of the method is higher,which proves the effectiveness and feasibility of the model.
作者
王樱达
丁泽
王延瓒
刘会永
张松
王佳宁
Wang Yingda;Ding Ze;Wang Yanzan;Liu Huiyong;Zhang Song;Wang Jianing(不详;Weichai Power Co.,Ltd.,Weifang,Shandong 261000,China)
出处
《工具技术》
北大核心
2023年第6期133-137,共5页
Tool Engineering
关键词
贝叶斯优化
双向长短时记忆网络
特征筛选
刀具磨损状态监测
bayesian optimization
bidirectional long short-term memory network
feature screening
tool wear detection
