基于K-means退化识别和随机森林的滚动轴承寿命分析

Life Analysis of Rolling Bearing Based on K-means Degradation Identification and Random Forest

针对滚动轴承剩余寿命(remaining useful life,RUL)预测方法中的寿命标签难以反映实际情况和预测精度不高的问题,提出了一种基于PCA-Kmeans-RF的滚动轴承剩余寿命分析方法。首先,通过小波降噪与快速傅里叶变换(fast fourier transform,FFT)分别对原始的水平振动信号进行降噪处理和提取多种时域、频域的特征;其次,通过方差过滤(variance filtering,VF)和主成分分析(principal component analysis,PCA)进行特征处理来选择更优的退化特征并融合特征构建健康指标(health indicators,HI);随后,利用K-means聚类确定样本的退化点和失效点,并以此构建剩余寿命的更为真实的退化标签,并将健康指标和时间标签输入到随机森林中进行训练和测试;最后,通过所提出的方法对IEEE PHM 2012滚动轴承数据集进行验证和对比。结果表明该方法取得了良好的结果。

The problem that the life label in the prediction method of residual useful life(RUL)of rolling bearing is difficult to reflect the actual situation and the prediction accuracy is not high,a analysis method of residual life of rolling bearing based on PCA-Kmeans-RF is proposed.Firstly,the original horizontal vibration signal is denoised by wavelet denoising and fast fourier transform(FFT),and various features in time domain and frequency domain are extracted.Secondly,through variance filtering(VF)and principal component analysis(PCA)for feature processing,better degraded features are selected and fused to construct health indicators(HI).Then,K-means clustering is used to determine the degradation point and failure point of the sample,and a more realistic degradation label of the remaining life is constructed,and the health index and time label are input into the random forest for training and testing.Finally,the IEEE PHM 2012 rolling bearing data set is verified and compared by the proposed method.The results show that this method has achieved results.