摘要
为了实现对滚动轴承故障位置和损伤程度的准确定位,将类别判别信息引入到无监督的稀疏编码中,提出一种有监督稀疏编码(Supervised Sparse Coding,SSC)方法,建立基于希尔伯特黄变换(Hilbert-Huang Transform,HHT)和SSC的振动信号特征提取和故障状态精细分类模型。首先,通过HHT获取振动信号的边际谱,然后,利用SSC为边际谱信息建立统一的字典库,并完成对边际谱的稀疏表示,实现干扰信息的滤除和故障目标敏感特征的二次提取,最后,使用SSC得到的稀疏系数完成对支持向量基(Support Vector Machine,SVM)分类器的训练。采用SKF-6205-2RS轴承试验台数据对提出方法进行实验分析,使用HHT-SSC-SVM模型,驱动端轴承故障状态识别率为99.5%,风扇端轴承故障状态识别率为98.25%,与文中其他模型相比,在故障状态识别率上有所提高,并且表现出来较强的适应能力。
A Supervised Sparse Coding (SSC) method was proposed by introducing classification discriminant analysis into unsupervised sparse encoding. Combined with Hilbert-Huang Transform (HHT), SSC and Support Vector Machine (SVM), a precise rolling bearing fault diagnosis method was proposed to effectively diagnose the position and damage degree of the bearing fault simultaneously. Because the HHT marginal spectrum of vibration signal contains a large number low correlation characteristic information for fault diagnosis, the unified dictionary library which can describe the feature of the marginal spectrum was established and the sparsity coefficients were calculated. At last, the sparsity coefficients were used as the fault characteristic vectors to train SVM classifier. SKF-6205-2RS rolling bearing fault data experiment demonstrated that the proposed method could accurately and effectively classify the position and damage degree of bearing fault state. Based on the vibration signal acquired by sensor located upon the motor drive side, the drive end bearing and fan end bearing fault state recognition rate were 99.5% and 98.25% respectively. © 2015, China Coal Society. All right reserved.
出处
《煤炭学报》
EI
CAS
CSCD
北大核心
2015年第11期2587-2595,共9页
Journal of China Coal Society
基金
国家重点基础研究发展计划(973)资助项目(2014CB046300)
关键词
希尔伯特黄变换
稀疏编码
支持向量机
特征提取
滚动轴承故障
Bearings (machine parts)
Codes (symbols)
Digital storage
Discriminant analysis
Electric drives
Failure analysis
Fault detection
Feature extraction
Mathematical transformations
State estimation
Support vector machines
