Generally, the transient characteristics of early bearing failure are not obvious. How to extract weak transient features is a big challenge. Dictionary learning has been successfully used to extract bearing fault features. However, the traditional dictionary learning is easy to fall into local optimum and cannot extract fault features from complex signals. And it often consumes huge computational costs. In order to solve the above problems, this paper proposes a fault pulse extraction and feature enhancement method for bearing fault diagnosis. Firstly, the bearing vibration signal is segmented in the time domain. Then this paper proposes a multi-scale alternating direction multiplier method for dictionary learning (MADMMDL) to extract fault impact signal from the segment signal. Finally, frequency spectrum averaging is used to enhance the bearing fault characteristic frequency. Through numerical simulation and rail transit transmission failure simulation experimental analysis, the feasibility of this method in bearing fault diagnosis is verified.

一般来说，轴承早期失效的瞬态特征不明显。如何提取弱瞬态特征是一个很大的挑战。字典学习已成功用于提取轴承故障特征。然而，传统的字典学习容易陷入局部最优，无法从复杂信号中提取故障特征。并且它通常会消耗巨大的计算成本。针对上述问题，本文提出了一种轴承故障诊断的故障脉冲提取与特征增强方法。首先，在时域中对轴承振动信号进行分段。然后本文提出了一种用于字典学习的多尺度交替方向乘法器方法（MADMMDL）从片段信号中提取故障影响信号。最后，频谱平均用于提高轴承故障特征频率。通过数值模拟和轨道交通传输故障仿真实验分析，验证了该方法在轴承故障诊断中的可行性。