Abstract: In the automatic cutting processes， accurate and reliable detecting of tool wear states was the key to ensure the processing quality and efficiency. Aiming at the problems of tedious feature extraction with low accuracy and traditional deep learning network could not extract the hidden information of data comprehensively， an improved model was proposed based on integration of CNN and BiLSTM by adding batch standardization layer to CNN and using two BiLSTM layers. The model could automatically extract the spatial and temporal features of cutting forces， vibration and sound signals after wavelet threshold denoising and down sampling to realize tool wear detection. Compared with CNN-BiLSTM model and traditional deep learning model， the accuracy and stability of the improved model are greatly improved. The proposed method provides technical support for accurately detecting the tool wear states in the automatic machining processes and improves the production efficiency and machining quality. 

Key words: wavelet threshold denoising, convolutional neural network（CNN）, bidirectional long short-term memory（BiLSTM） network, tool wear detection

摘要： 自动化切削加工过程中，准确可靠地监测刀具磨损状态是保证加工质量和加工效率的关键。针对刀具磨损状态相关特征提取繁琐、准确率低及传统的深度学习网络不能全面提取数据隐含信息等问题，提出了一种以卷积神经网络（CNN）和双向长短时记忆（BiLSTM）网络集成模型为基础并通过在卷积神经网络中添加批量标准化层和采用两个双向长短时记忆网络层的改进模型，该模型通过自动提取小波阈值降噪等预处理和降采样后的切削力、振动和声音信号的空间和时序特征来实现刀具磨损状态监测。将改进模型与CNN-BiLSTM模型及传统的深度学习模型进行对比，发现改进模型在精度和稳定性方面有较大提升。所提方法为准确监测自动化加工过程中刀具磨损状态、提高生产效率和加工质量提供了技术支持。

关键词: 小波阈值降噪, 卷积神经网络, 双向长短时记忆网络, 刀具磨损状态监测