【摘要】 机械故障特征具有周期性、稀疏性以及被噪声污染严重特点,而大部分特征抽取方法(如局部线性嵌入(locally linear embedding,LLE)、局部切空间排列(local tangent space alignment,LTSA))性能往往受到噪声影响.因此,抑制振动信号噪声、抽取有效故障特征成为机械故障检测的关键.本文提出融合奇异值分解与周期重叠簇稀疏(reweighted singular value decomposition integrating with periodic overlapping group sparsity,RSVD-POGS)的机械故障稀疏特征抽取方法.该方法首先利用RSVD把多成分振动信号分解为奇异成分集合,并使用周期调制强度(periodic modulation intensity,PMI)准则选择有效奇异成分,然后使用POGS从奇异成分提取稀疏周期冲击特征,并由选择的奇异成分重构原始信号,增强周期稀疏故障信号特征.最后,使用低SNR仿真周期冲击信号对RSVD-POGS算法与POGS方法进行对比,并将RSVD-POGS方法应用于实验台轴承正常和故障信号的特征提取中.实验结果表明,该算法可以有效地提取稀疏微弱故障特征,具有较大的优越性.更多还原

【Abstract】 The machinery fault features are generally periodic,sparse and corrupted by heavy background noise.Most feature extraction methods like locally linear embedding(LLE),local tangent space alignment(LTSA) and etc.,are susceptible to noise.Therefore,signal denoising and faulty feature enhancement are critical for machinery fault detection.To address the issue,a reweighted singular value decomposition integrating with periodic overlapping group sparsity(RSVD-POGS)-based feature extraction approach is developed for machinery fault detection in this study.Firstly,RSVD is introduced to decompose multiple-component vibration signal into a set of singular components and periodic modulation intensity(PMI) criterion is utilized to select salient singular components.POGS is then conducted to enhance sparse and periodic impulse features hidden in the remained singular components,and the original signal can be constructed by the denoised singular components.The comparison between RSVD-POGS and POGS is carried out on low SNR simulated sparse periodic signal and RSVD-POGS is utilized to extract features rom normal and fault vibration signals collected from experimental platform.The experimental results demonstrate that the proposed method can effectively enhance the periodic and sparse fault features for machinery fault diagnosis.更多还原