基于数据融合LSSVM的滚动轴承剩余寿命预测

发布时间：2018-04-21 01:28

本文选题：相关系数 + 提升小波变换　；参考：《西南交通大学》2017年硕士论文

【摘要】：轴承是工业设备的重要连接部件,一直以来,滚动轴承都是设备故障状态的热门研究对象。研究滚动轴承的剩余寿命有助于提高机械设备的使用寿命,提前制定合理的故障维护措施,大大的降低轴承故障给企业带来的经济损失和意外伤害。本论文分别采用最小二乘支持向量机(Least Squares Support Vector Machine,以下简称LS-SVM)回归预测模型进行全寿命滚动轴承的剩余寿命预测,以马氏距离与核主成分分析融合得到的特征作为研究对象,经过预测结果对比,核主成分分析融合的特征预测效果更好。在信号分析与处理中,需要对采集的数据进行预处理。在本论文中,我们采用改进的基于提升小波变换(Lifting Wavelet Transform,LWT)方法进行滚动轴承振动信号的降噪处理。首先,对全寿命数据进行提升小波分析得到分解后的小波系数,然后对小波系数进行提升小波逆变换得到重构之后的信号,通过计算信号的重构分量与原信号的相关系数(Correlation Coefficient,CC),对小于设定阈值的小波系数置零,最后再使用处理后的小波系数进行提升小波重构以完成消噪处理。经过预处理的数据需要进行特征提取,研究选用时域特征、频域特征和小波特征作为表征信号特性的参数。在模型建立之前,需要利用提取得到的信号特征构造模型的输入特征参数。第四章使用马氏距离(Mahalanobis Distance,MD)与核主成分分析(Kernel Principal Component Analysis,KPCA)的方法进行特征参数的融合,得到两组不同的信号特征,即分别为单参数特征和多参数特征。第五章主要是研究LS-SVM模型的建立和滚动轴承剩余寿命的预测。选择径向基函数作为模型的核函数,通过参数优化得到预测效果更好的惩罚因子与核函数参数,进而得到LS-SVM的模型。论文最后利用LS-SVM模型对单参数输入与多参数输入的滚动轴承的剩余寿命进行预测。试验研究结果表明,基于核主成分分析(KPCA)原理进行特征融合得到的多参数输入的LS-SVM模型的寿命预测效果更好,精度更高,其在实际工程应用和科学研究中具有更重大的意义。
[Abstract]:Bearing is an important connecting part of industrial equipment. Rolling bearing is the hot research object of equipment fault state all the time. The study of the residual life of rolling bearing is helpful to improve the service life of machinery and equipment, make reasonable maintenance measures in advance, and greatly reduce the economic loss and accidental injury caused by bearing failure. In this paper, the least squares support vector machine (LS-SVM) regression model is used to predict the residual life of rolling bearings. The features obtained from the fusion of Markov distance and kernel principal component analysis (KPCA) are taken as the research objects. By comparing the prediction results, the feature prediction effect of kernel principal component analysis fusion is better. In signal analysis and processing, the collected data need to be preprocessed. In this paper, an improved lifting Wavelet transform method based on lifting wavelet transform is used to reduce the noise of rolling bearing vibration signal. Firstly, the decomposed wavelet coefficients are obtained by lifting wavelet analysis to the whole life data, and then the reconstructed signals are obtained by lifting wavelet inverse transform of wavelet coefficients. By calculating the correlation coefficient of the reconstructed component of the signal and the correlation coefficient of the original signal, the wavelet coefficients less than the set threshold are set to zero. Finally, the wavelet coefficients after processing are reconstructed by lifting the wavelet coefficients to complete the denoising process. The preprocessed data need to be extracted by feature extraction. The time domain feature, frequency domain feature and wavelet feature are selected as the parameters to characterize the signal characteristics. Before the model is established, the input feature parameters of the model need to be constructed by using the extracted signal features. In chapter 4, the method of Mahalanobis distance MD) and kernel principal component analysis (Kernel Principal Component Analysis) are used to fuse the feature parameters, and two sets of different signal features are obtained, that is, single parameter feature and multi-parameter feature respectively. The fifth chapter mainly studies the establishment of LS-SVM model and the prediction of the remaining life of rolling bearing. The radial basis function is chosen as the kernel function of the model, and the penalty factor and kernel function parameter with better prediction effect are obtained by parameter optimization, and then the model of LS-SVM is obtained. Finally, LS-SVM model is used to predict the residual life of rolling bearing with single parameter input and multi parameter input. The experimental results show that the multi-parameter input LS-SVM model based on the kernel principal component analysis (KPA) principle has better prediction effect and higher precision, and it has more significance in practical engineering application and scientific research.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH133.33

【参考文献】