基于一维卷积长短时记忆网络的多信号融合刀具磨损评估

Multisignal Fusion Assessment of Tool Wear Based on 1DCNN-LSTM

为实现刀具磨损状态的有效在线监测,提出一种基于一维卷积长短时记忆网络的多信号融合刀具磨损评估模型。该模型综合使用加工过程中主轴和工作台的振动和声发射信号,以实现信号间的优势互补,弥补单一信号的不足;基于一维卷积的特征学习能力和长短时记忆网络的时序特征分析能力,充分挖掘信号中包含的刀具磨损状态信息;最后通过全连接层和softmax分类器对刀具磨损状态进行评估。试验结果表明,该模型在各单一工况下对刀具磨损状态的识别准确率均可达93.8%以上,整体工况下识别准确率达95.3%,具有很好的稳定性和多工况通用性。

In order to effectively monitor the tool wear condition,a multisignal fusion tool wear evaluation model based on 1DCNN-LSTM is proposed.The vibration signal and acoustic emission signal of the spindle and working table are utilized to realize advantageous complementarities between signals to offset the defects of single signal;1D convolution and LSTM are applied to adaptively extract features and temporal properties of signals to obtain more information relative to tool wear condition;fully connected layers and the softmax classifier are used to evaluate the stage of tool wear finally.The experiment indicates that the proposed 1DCNN-LSTM model can acquire accuracy of more than 93.8%under each signal working condition and accuracy of 95.3%under overall working condition,which has great stability and versatility.