中文摘要：

针对强背景噪声下机械故障微弱暂态特征表示和有效提取的难题,提出了通用的稀疏优化特征提取算法。算法针对含噪声冲击性特征提取问题设计了稀疏优化表征函数,该函数融合了冲击特征的保真度与惩罚函数因子,考虑了正则化参数以适应不同工程背景下各分析因子的实际影响,实现处理结果稀疏性极大化。同时,引入受控极小化方法对设计的表征函数进行转化,分解成一系列凸优化分析问题。提出了针对离散信号的有限差分式数值迭代算法,验证了其快速收敛性和数值稳定性,提出的算法对机械故障诊断的数字采样信号具有普遍适用性。将所提出的算法应用于实验室环境下的轴承故障特征识别中,无论是低噪声还是低信噪比白噪声环境下,振动信号中的冲击特征都得到了显著增强,在Hilbert包络谱中的故障特征频率及其高次谐波比能量中占优。所提出的算法还应用于电力机车走行部轮对的故障诊断中,在高强度的工程有色噪声环境下精确提取了其中的冲击衰减成分,在时域和频域诊断结果中都得到了准确的验证,指导了诊断实践。

英文摘要：

To deal with the effective representation and extraction of incipient transient features of mechanical fault,ageneral sparsity based identification approach is proposed.This approach designs a sparsity optimization function that integrates impulsive feature preserving factor and penalty function factor,and takes the regularization parameter into consideration such as to address the actual factor weights in different situations.The majorization minimization is introduced to simplify the designed function into a series of convex optimization problems.A finite difference based numerical iterative method is developed for the proposed approach,and its fast convergence and numerical stability are illustrated.The proposed approach is versatile to digital signal processing of mechanical fault detecting practices,and is applied to bearing fault identifications in lab.It is shown that no matter in high or low noise backgrounds,the impulsivecomponents are significantly enhanced,which can be verified in the dominant energy ratios of characteristic frequencies in the Hilbert envelope spectrum.This approach is also utilized to conduct bearing fault diagnosis of traction part of electrical locomotive,and the impulsive features masked by heavy colored noises are effectively detected in time domain and spectral domain.