结合连续小波变换和多约束非负矩阵分解的故障特征提取方法  

Fault feature extraction method combining continuous wavelet transformation with multi-consitraint nonnegative matrix factorization

作　　者：唐曦凌 梁霖[2] 高慧中[2] 罗爱玲[2] 

机构地区：[1]广东电网公司电力科学研究院,广州510080 [2]西安交通大学机械工程学院,西安710049

出　　处：《振动与冲击》2013年第19期7-11,共5页Journal of Vibration and Shock

摘　　要：为了在故障早期从信噪比较低的振动信号中提取故障特征,提出了一种结合小波变换和多约束非负矩阵分解振动信号特征提取方法。首先,采用最小小波熵测量提取出最优时频系数矩阵。然后,根据故障特征在系数矩阵中的表现规律,采用基于稀疏性和光滑性约束的非负矩阵分解算法对小波系数矩阵进行非线性降维,从而提取信噪比较高的故障特征。最后,通过仿真数据和实际数据对该方法进行了验证,结果表明该方法能够在时域中提取出微弱的故障特征,实现机械状态的早期故障诊断。In order to extract weak fault features from vibration signals mixing with noise, a method combining continuous wavelet transformation with multi-constraint nonnegative matrix factorization was presented here. Firstly, the continuous wavelet transformation was adopted to transform a signal into a time-frequency domain, and the optimal time-frequency coefficient matrix was extracted out using the minimum wavelet entropy method. Secondly, the non-negative matrix factorization based on sparsity and smoothness of constraints was used to reduce the dimensions of the optimal coefficient matrix according to the fault characteristics shown in the matrix so as to extract fault features with a higher signal-noise ratio. Finally, simulations and test data showed that this approach can effectively extract weak fault features and realize the early fault diagnosis.

关 键 词：特征提取 连续小波变换 非负矩阵分解 故障诊断 

分 类 号：TH165.3[机械工程—机械制造及自动化]