Abstract: To improve the accuracy and generalization,a tool condition monitoring approach is proposed based on convolutional neural network(CNN).To prevent the loss of signal information caused by the data preprocessing,signals in time domain were used to analyze tool wear condition quantitatively.Instead of manually extracting features from signals,an adaptive method is developed by using deep network.In order to mine tiny features,deeper neural network is used.The experiment study verifies the approach's excellent performance.Both accuracy and generalization are improved,when the limitation of manual feature extraction is avoided.The comparison with relevant studies also validates its feasibility and efficiency.

Key words: tool condition monitoring, tool wear condition, signals in time domain, convolution neural network, feature extraction

摘要： 为了提高刀具磨损在线监测的精度和泛化性能,提出一种基于卷积神经网络的刀具磨损量在线监测模型。利用时域传感器信号对刀具磨损量进行定量分析,避免数据预处理带来的信息丢失;采用深度网络自适应地提取特征,取代传统的人工特征提取过程,并通过加深网络进一步挖掘信号中隐藏的微小特征。实验结果表明,该模型对刀具后刀面磨损量监测效果较好,可以有效避免人为特征提取的局限,精度和泛化性都有一定程度的提高。与相关研究的对比也证实了其可行性和有效性。

关键词: 刀具状态监测, 刀具磨损量, 时域传感器信号, 卷积神经网络, 特征提取