基于深度SSDAE网络的刀具磨损状态识别  

Tool Wear State Recognition Based on Deep Stacking Sparse Denoising Auto-encoder Network

作　　者：郭润兰[1] 尉卫卫 王广书 黄华[1] GUO Runlan;YU Weiwei;WANG Guangshu;HUANG Hua(School of Electrical and Mechanical Engineering,Lanzhou University of Technology Lanzhou,730050,China)

机构地区：[1]兰州理工大学机电工程学院,兰州730050

出　　处：《振动．测试与诊断》2024年第2期305-312,410,411,共10页Journal of Vibration,Measurement & Diagnosis

基　　金：国家自然科学基金资助项目(52365057,51965037)。

摘　　要：针对刀具磨损状态识别过程中采集数据量大、干扰信号复杂且需人为选择特征参数的问题,为提高刀具磨损状态识别模型的鲁棒性与泛化性,提出了一种数据驱动下深度堆叠稀疏降噪自编码(stacking sparse denoising auto-encoder,简称SSDAE)网络的刀具磨损状态识别方法,实现隐藏在数据中深层次的数据特征自动挖掘。首先,将原始振动信号分解为一系列固有模态分量(intrinsic mode function,简称IMF),并采用皮尔逊相关系数法选取了最优固有模态来组合一个新的信号;其次,采用SSDAE网络自适应提取特征后对刀具磨损阶段进行了状态识别,识别精度达到98%;最后,对网络模型进行实验验证,并与最常用的刀具磨损状态识别方法进行了对比。实验结果表明,所提出的方法能够很好地处理非平稳振动信号,对不同刀具磨损阶段状态的识别效果良好,并具有较好的泛化性能和可靠性。In order to improve the robustness and generalization of the tool wear status recognition model,a data-driven tool wear status recognition method with deeply stacking sparse denoising auto-encoder(SSDAE)network is proposed to achiev-e automatic mining of data features hidden in the data at a deep level.First,the original vibration signal is decomposed a series of intrinsic mode function(IMF).The Pearson correlation coefficient method is used to select the optim-al intrinsic mode function to combine a new signal.Secondly,the SSDAE adaptive feature extraction is used to ident-ify the state of the tool wear stage,and the accuracy of the tool wear state identification reached 98%.Finally,the netwo-rk model is experimentally validated and compared with the most commonly used tool wear state recognition methods.The experimental results show that the proposed method can handle non-smooth vibration signals well.Therefore,this method has good recognit-ion ef-fect on different tool wear stage states good generalization performance and high reliability.

关 键 词：深度堆叠稀疏自编码网络 变分模态分解 K-最近邻分类器 自适应特征提取 状态识别 

分 类 号：TH164[机械工程—机械制造及自动化] V262.33[航空宇航科学与技术—航空宇航制造工程]