基于小波包和SVM一类学习在电机异音检测中的应用  

Application of the Motor Abnormal Sound Detection Based on Wavelet Packet and SVM One-Class Learning

作　　者：温浩[1] 刘力源[1] 

机构地区：[1]五邑大学信息工程学院,广东江门529020

出　　处：《测控技术》2015年第2期35-38,43,共5页Measurement & Control Technology

摘　　要：生产线上大批量小型电机普遍通过训练有素的工人进行音质检测。针对电机声音信号的统计特性及其人工检测特点,采用小波包变换对电机的声信号进行分解,并提取其频带系数的奇异值作为特征,映射到特征矢量所张成的状态空间。考虑到生产线上异音样本量少、获取困难、个体差异造成异音等问题难以分析,引入支持向量机一类学习算法进行异音检测。通过对电机声信号的实测数据进行分析,充分利用小波包优良的时频局部化特性和支持向量机在小样本情况下出色的学习性能及全局最优能力,验证这种方法的有效性。A large number of small motors are detected in production line by trained workers for abnormal sound.According to the statistical characteristics of the motor sound signal and artificial detection,the acoustic signal of motor is decomposed by the method of wavelet packet transform,and its singular value is extracted as the characteristics of frequency band coefficients to be mapped to the state space generated by the feature vector.Because it is difficult to analyze the less abnormal noise motor samples,difficult acquirement and abnormal noise caused by the indicidual differences in production line,the support vector machine（SVM） one-class learning algorithm is used to detect the abnormal sound.Through analyzing the measured motor sound signals and making full use of the excellent time-frequency localization characteristics of wavelet packet and the excellent study performance and global optimal ability of SVM under the condition of small sample,the effectiveness of this approach is verified.

关 键 词：异音检测 小波包分解 奇异值 支持向量机 一类学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]