针对机械早期故障引起的冲击特征微弱,易受强背景信号和噪声的干扰而难以提取的问题,提出一种奇异值分解(Singular Value Decomposition,SVD)差分谱与S变换相结合的微弱冲击特征提取方法。将原始信号构造成Hankel矩阵,采用SVD对重构矩...针对机械早期故障引起的冲击特征微弱,易受强背景信号和噪声的干扰而难以提取的问题,提出一种奇异值分解(Singular Value Decomposition,SVD)差分谱与S变换相结合的微弱冲击特征提取方法。将原始信号构造成Hankel矩阵,采用SVD对重构矩阵进行分解;利用奇异值差分谱确定降噪阶次进行降噪;采用S变换对降噪后的信号进行时频分析,提取信号中的微弱冲击特征信息。通过数值仿真和实际轴承故障数据的对比,表明该方法可有效辨别轴承振动信号中故障引起的早期微弱冲击特征,为轴承故障诊断提供先验信息。展开更多
The existence of remnant particles, which significantly reduce the reliability of relays, is a serious problem for aerospace relays. The traditional method for detecting remnant particles-particle impact noise detecti...The existence of remnant particles, which significantly reduce the reliability of relays, is a serious problem for aerospace relays. The traditional method for detecting remnant particles-particle impact noise detection （PIND）-can be used merely to detect the existence of the particle; it is not able to provide any information about the particles＇ material. However, information on the material of the particles is very helpful for analyzing the causes of remnants. By analyzing the output acoustic signals from a PIND tester, this paper proposes three feature extraction methods： unit energy average pulse durative time, shape parameter of signal power spectral density （PSD）, and pulse linear predictive coding coefficient sequence. These methods allow identified remnants to be classified into four categories based on their material. Furthermore, we prove the validity of this new method by processing P1ND signals from actual tests.展开更多
文摘针对机械早期故障引起的冲击特征微弱,易受强背景信号和噪声的干扰而难以提取的问题,提出一种奇异值分解(Singular Value Decomposition,SVD)差分谱与S变换相结合的微弱冲击特征提取方法。将原始信号构造成Hankel矩阵,采用SVD对重构矩阵进行分解;利用奇异值差分谱确定降噪阶次进行降噪;采用S变换对降噪后的信号进行时频分析,提取信号中的微弱冲击特征信息。通过数值仿真和实际轴承故障数据的对比,表明该方法可有效辨别轴承振动信号中故障引起的早期微弱冲击特征,为轴承故障诊断提供先验信息。
基金China Science Technology and Industry Foundation for National Defense (FEBG 27100001)
文摘The existence of remnant particles, which significantly reduce the reliability of relays, is a serious problem for aerospace relays. The traditional method for detecting remnant particles-particle impact noise detection （PIND）-can be used merely to detect the existence of the particle; it is not able to provide any information about the particles＇ material. However, information on the material of the particles is very helpful for analyzing the causes of remnants. By analyzing the output acoustic signals from a PIND tester, this paper proposes three feature extraction methods： unit energy average pulse durative time, shape parameter of signal power spectral density （PSD）, and pulse linear predictive coding coefficient sequence. These methods allow identified remnants to be classified into four categories based on their material. Furthermore, we prove the validity of this new method by processing P1ND signals from actual tests.
