文章摘要
龚子维,刘希强,张仲宁,杨京,程建春,刘翔雄.基于VMD-PNN的砂轮钝化声发射检测[J].声学技术,2021,40(2):260~268
基于VMD-PNN的砂轮钝化声发射检测
Acoustic emission detection of blunting states of grinding wheel based on VMD-PNN
投稿时间:2020-03-09  修订日期:2020-03-26
DOI:10.16300/j.cnki.1000-3630.2021.02.018
中文关键词: 砂轮钝化  声发射  变分模态分解(VMD)  概率神经网络(PNN)
英文关键词: grinding wheel blunting  acoustic emission  variational mode decomposition (VMD)  probabilistic neural network (PNN)
基金项目:国家自然科学基金项目(11374157)。
作者单位E-mail
龚子维 南京大学声学研究所, 江苏南京 210093
人工微结构科学与技术协同创新中心, 江苏南京 210093 
 
刘希强 南京大学声学研究所, 江苏南京 210093
人工微结构科学与技术协同创新中心, 江苏南京 210093 
 
张仲宁 南京大学声学研究所, 江苏南京 210093
人工微结构科学与技术协同创新中心, 江苏南京 210093 
 
杨京 南京大学声学研究所, 江苏南京 210093
人工微结构科学与技术协同创新中心, 江苏南京 210093 
yangj@nju.edu.cn 
程建春 南京大学声学研究所, 江苏南京 210093
人工微结构科学与技术协同创新中心, 江苏南京 210093 
 
刘翔雄 华辰精密装备(昆山)股份有限公司, 江苏昆山 215337  
摘要点击次数: 621
全文下载次数: 286
中文摘要:
      在磨削加工过程中,加工刀具即砂轮会发生钝化现象,砂轮表面磨损影响加工精度和工件质量,需要及时检测并修整。磨粒的塑性变形、破碎、断裂等会产生声发射信号,能够作为精确识别砂轮钝化状态的依据,且不易被噪声干扰,因此提出一种基于变分模态分解(Variational Mode Decomposition,VMD)和概率神经网络(Probabilistic Neural Network,PNN)的砂轮钝化声发射检测方法。VMD可以将原始信号分解为多个本征模态函数(Intrinsic Mode Function,IMF)分量,筛选其中峭度较大的分量重构即得到声发射信号。声发射检测的关键是特征参数的选取,在相关研究基础上本文提出了声发射包络能量占比作为一个重要的特征参数,并选取了共5种特征参数,构建出5维特征向量数据集,输入到PNN中进行训练,经过测试识别准确度达到94.5%。该方法建立了声发射信号特征参数与砂轮不同钝化状态的关系,能够对砂轮严重钝化状态给出准确预警,具有实际应用价值。文章比较了声发射信号不同特征参数用于识别砂轮钝化状态的准确度,对特征参数的选用具有参考意义。
英文摘要:
      During the grinding process, the blunting phenomenon occurs on the processing tool, i.e. the grinding wheel. The wear of the grinding wheel surface affects the machining accuracy and the quality of the workpiece, and it needs to be detected and repaired in time. The plastic deformation, fragmentation, and fracture of the abrasive particles will generate acoustic emission (AE) signals, which can be used as a basis identifying the blunting states of the grinding wheel, and it is not easy to be disturbed by noise. Therefore, an AE detection method of grinding wheel blunting states based on variational mode decomposition (VMD) and probabilistic neural network (PNN) is proposed. VMD can decompose the original signal into multiple intrinsic mode function (IMF) components, and filter out the components with larger kurtosis to reconstruct AE signal. The key to AE detection is the selection of characteristic parameters. Based on the related researches, the proportion of envelope energy is presented as an important characteristic parameter, and a total of 5 characteristic parameters are selected to construct a five-dimensional characteristic vector dataset and input to PNN for training. After testing, the recognition accuracy reaches 94.5%. This method establishes the relationship between the characteristic parameters of AE signal and the different blunting states of grinding wheel, which can accurately predict the severe blunting state of grinding wheel and has practical application value. Moreover, the accuracies using different characteristic parameters of AE signals to identify the blunting states of grinding wheel are compared in this paper, which has reference significance for the selection of characteristic parameters.
查看全文   查看/发表评论  下载PDF阅读器
关闭