基于多源同步信号与深度学习的刀具磨损在线识别方法

为提高刀具状态监测系统的实用性、避免实际加工过程中工序变换产生的信号干扰，提出一种基于多源同步信号与深度学习的刀具磨损在线识别方法。该方法利用自动触发的方式实现了机床运行在特定工序时的刀具振动、主轴功率、数控系统参数等多源信号的同步在线采集，保证信号同步性的同时有效避免了因工序变换而产生的信号波动干扰；进一步利用高频振动特征实现了 “切削过程”与“切削间隙”采集样本的准确划分，并基于皮尔逊积矩相关系数筛选出强关联特征，保证了多源监测信号融合样本的可用性；最后基于一维卷积神经网络建立了刀具磨损在线识别模型。实验结果表明，该方法无论从识别精度还是诊断效率，均能实现实际加工过程中刀具磨损状态的在线识别。

Tool Wear Online Recognition Method Based on Multi-source Synchronous Signals and Deep Learning

To improve the practicability of the tool condition monitoring systems and avoid signal interference caused by operation changes in the actual machining processes, a tool wear online recognition method was proposed based on multi-source synchronous signals and deep learning. The proposed method leveraged the automatic triggering to achieve the synchronous online collection of multi-source signals, i.e., tool vibration, spindle power, and CNC system parameters while the machine tool was running in a specific process. The synchronization of the collected signals might be ensured and the signal fluctuation interference caused by operation changes might be also avoided. Moreover, high-frequency vibration characteristics were employed to achieve the accurate division of samples collected during the “cutting process” and the “cutting interval”, and the Pearson correlation coefficient was utilized to extract the highly correlated features, guaranteeing the availability of the multi-source signal fusion samples. Finally, a tool wear online recognition model was established based on the one-dimensional convolutional neural network. The experimental results indicate that the proposed method fulfills the requirements of online recognition of tool wear in the actual machining processes in terms of recognition accuracy and diagnostic efficiency.