基于深度门控循环单元神经网络的刀具磨损状态实时监测方法  

Real-time monitoring method for wear state of tool based on deep bidirectional GRU model

作　　者：陈启鹏[1] 谢庆生[1] 袁庆霓[1] 黄海松[1] 魏琴[2] 李宜汀 CHEN Qipeng;XIE Qingsheng;YUAN Qingni;HUANG Haisong;WEI Qin;LI Yiting(Key Laboratory of Advanced Manufacturing Technology,Ministry of Education,Guizhou University,Guiyang 550025,China;State Key Laboratory of Public Big Data,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学现代制造技术教育部重点实验室,贵州贵阳550025 [2]贵州大学公共大数据国家重点实验室,贵州贵阳550025

出　　处：《计算机集成制造系统》2020年第7期1782-1793,共12页Computer Integrated Manufacturing Systems

基　　金：国家重点研发计划资助项目(2018YFB1004305);国家973计划资助项目(2015CB856001);国家自然科学基金资助项目(51865004);贵州省科技重大专项计划资助项目([2017]3004);贵州省科技厅科技资助项目([2017]2870);贵州省教育厅科技人才支持计划资助项目([2017]062)。

摘　　要：为监测生产加工过程中的刀具磨损状态,提出一种基于深度门控循环单元神经网络的轻量化状态监测模型。首先,预处理阶段对加速度传感器采集的时序信号进行小波阈值去噪,并将每次刀具进给产生的冗长信号划分为多个训练样本,以滤除噪声、改善算法的鲁棒性;然后,利用卷积神经网络(CNN)从时序信号输入中自适应地提取特征,构建深度双向门控循环单元(BiGRU)神经网络学习特征向量间的时序信息,并将Attention机制的思想引入其中,自适应地感知对磨损状态分类结果有关联的网络权重,并对其进行合理分配,避免因人工提取特征带来的复杂性和局限性。实验结果表明,所提方法能够对传感器采集的原始数据实时准确地预测刀具磨损状态,在识别精度和泛化能力上均达到了较好的效果,为实际工业场景下的刀具磨损状态监测提供了新的思路。To monitor the tool wear state of Computerized Numerical Control(CNC) machining equipment in real time in a manufacturing workshop, a lightweight real-time monitoring method based on deep gated recurrent unit neural network was proposed. The wavelet threshold denoising on the timing signals collected by the acceleration sensors was performed in the preprocessing stage, and the redundant signal generated by each tool feed was divided into a plurality of training samples to filter out noise and improve the robustness of the algorithm. Convolutional Neural Network(CNN) was used to extract deep features from time-series signal as an input, and then the deep Bidirectional Gated Recurrent Unit(BiGRU) neural network was constructed to learn the time-series information between the feature vectors. Attention mechanism was introduced to self-adaptively perceive the network weights associated with the classification results of wear state and distribute the weights reasonably. In this way, the limitations of traditional manual feature extraction steps could be avoided. The experimental results demonstrated that the proposed method could accurately predict the tool wear state in real time for raw data collected by the sensors, and achieve better results in recognition accuracy and generalization ability, which provided a new way to detect tool wear state in practical industrial scenarios.

关 键 词：刀具磨损状态 实时监测 小波去噪 卷积神经网络 双向门控循环单元 Attention机制 

分 类 号：TP391.7[自动化与计算机技术—计算机应用技术]