基于密度可调谱聚类的半监督SVM机械早期故障预示方法

发布时间：2018-03-23 04:11

本文选题：谱聚类　切入点：密度　出处：《华南理工大学》2011年硕士论文　论文类型：学位论文

【摘要】：由于机械(如汽车变速器)早期故障的特征信号微弱,容易淹没在强噪声之中,而且已知故障模式样本不足,传统的频谱分析方法对故障的早期检测不敏感,因此,开展早期故障智能预示的研究工作具有重要意义。本文提出了基于密度可调谱聚类的半监督SVM(DSTSVM)方法,利用基于密度可调谱聚类的思想对数据进行特征提取,并且构造半监督SVM(TSVM)的核函数,采用梯度下降法对TSVM进行协同训练,实现对数据的分类,通过仿真和实例,将该方法与SVM、TSVM和基于聚类核的半监督SVM(CKSVM)进行对比分析,证明该方法能有效反映数据结构信息,用少量已知标签样本便能得到较高分类正确率。利用传动试验台对汽车变速箱进行无故障、齿轮轻微点蚀和齿轮轻微剥落试验,通过时域、频域方法分析出早期故障诊断的困难所在,将基于密度可调谱聚类的半监督SVM方法应用到齿轮早期故障预示中,分别采用经过PCA选择的时域特征指标、构造的频域能量因子作为输入,并将多传感器数据进行融合学习,与其它方法进行对比,证明了该方法在齿轮故障预示中的有效性和优越性。采用美国西储大学的电机轴承故障数据,对内圈、外圈、滚动体故障做了时频域分析,分析出滚动体早期故障诊断的困难,采用SVM、TSVM、CKSVM和DSTSVM对滚动体故障进行检测,并且对四种模式进行了分类识别,验证了DSTSVM方法在轴承早期故障预示中的有效性。
[Abstract]:Because the characteristic signal of early fault of machinery (such as automobile transmission) is weak, easily submerged in strong noise, and the sample of known fault mode is insufficient, the traditional spectrum analysis method is not sensitive to the early detection of fault. It is of great significance to carry out the research on early fault intelligent prediction. In this paper, a semi-supervised SVM DST SVM method based on density tunable spectral clustering is proposed. The feature extraction of data based on density tunable spectrum clustering is used, and the kernel function of semi-supervised SVMtSVM is constructed, and the gradient descent method is used to train TSVM cooperatively. The classification of data is realized. Through simulation and example, the method is compared with SVMN TSVM and semi-supervised SVMN CKSVM based on clustering kernel. It is proved that this method can effectively reflect the information of data structure. A high classification accuracy can be obtained by using a small number of known tag samples. The transmission test bench is used to test the automobile gearbox without fault, the gears are slightly pitting and the gears are peeling off slightly. The difficulties of early fault diagnosis are analyzed by time-domain and frequency-domain methods. The semi-supervised SVM method based on density adjustable spectrum clustering is applied to the early fault prediction of gears. The time-domain characteristic index selected by PCA is used to construct the frequency-domain energy factor as input, and the multi-sensor data is fused to learn. Compared with other methods, this method is proved to be effective and superior in gear fault prediction. Based on the fault data of motor bearing from the University of Western Reserve of USA, the fault of inner ring, outer ring and rolling body is analyzed in time and frequency domain, and the difficulty of early fault diagnosis of rolling body is analyzed. The fault of rolling body is detected by SVM TSVM CKSVM and DSTSVM. The classification and recognition of four kinds of patterns are carried out to verify the effectiveness of DSTSVM method in early bearing fault prediction.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TH165.3

【参考文献】