基于主元分析法的切丝机声音信号特征提取  

Feature extraction of sound signal for wire cutting machine based on principal component analysis method

作　　者：钱杰 秦基伟 赵建波 葛立武 焦阳 吕烁屹 赵炳荣 QIAN Jie;QIN Jiwei;ZHAO Jianbo;GE Liwu;JIAO Yang;Lü Shuoyi;ZHAO Bingrong(Cigarette Factory of Hongyun Honghe Group,Kunming 650032,Yunnan,China)

机构地区：[1]红云红河集团昆明卷烟厂,云南昆明650032

出　　处：《农业装备与车辆工程》2025年第3期111-116,共6页Agricultural Equipment & Vehicle Engineering

摘　　要：为有效提取声音信号特征实现切丝机刀盘故障监测,提出一种以主元分析法(PCA)为依托的特征提取模式。针对Tobspin切丝机,分别采集处于正常工作及刀盘出现故障时发出的声音信号,使用Welch法对声音信号进行功率谱估计,采用主元分析法针对切丝机的声音信号进行功率谱估计后分段处理获取的基本特征实施特征提取,将原始的特征由300维降低到67维;使用支持向量机(SVM)分类模型对完成特征提取后的样本集合进行验证,分类模型的准确率由57.14%提高至85.71%,模型运算时间由0.97 s下降至0.53 s。所提方法使特征提取后的特征分布更加确定、随机性更小,减少了特征的维数及冗余度,可作为切丝机声音信号分类模型的输入,实现切丝机的刀盘故障识别。A feature extraction mode based on principal component analysis(PCA) was proposed in order to effectively extract sound signal features and realize cutter head fault monitoring.For Tobspin cutter,the sound signals emitted when the cutter head was in normal operation and when the cutter head was faulty were collected respectively.Welch method was used to estimate the power spectrum of the sound signals.Principal component analysis was used to estimate the power spectrum of the sound signals of the cutter.Reducing the original feature from 300 to 67 dimensions;support vector machine(SVM) classification model was used to verify the sample set after feature extraction.The accuracy of the classification model was improved from 57.14% to 85.71%,and the operation time of the model was reduced from 0.97 s to 0.53 s.The proposed method makes the feature distribution after feature extraction more certain and less randomness,reduces the dimension and redundancy of features,and can be used as the input of sound signal classification model to realize cutter head fault identification.

关 键 词：特征提取 声音信号 主元分析法 切丝机 支持向量机 

分 类 号：TS43[农业科学—烟草工业] TP277[自动化与计算机技术—检测技术与自动化装置]