Abstract: Due to effects of the non-stationary operations, noisy working environment and strong electromagnetic interference for the wind turbines, the fault impulses of rolling bearings were submerged by strong noise, and the weak features were difficult to accurately identify. To solve the above problems, a frequency domain group sparse model with graph regularization constraints was proposed, which might effectively extract fault features of rolling bearings without periodic prior knowledge. Firstly, vibration signals were converted into graph signals to construct the graph regularization constraints, and the structured information was utilized to guide the penalty strength to improve the accuracy of sparse reconstruction. Secondly, the frequency domain group sparse model with graph regularization constraints was constructed, the method was given to determine the shrinkage threshold of the in-group components, and the objective function was simplified with the proximal mapping to optimize the solution. Finally, the parameters of the model were optimized by using the constructed comprehensive index and the moth flame optimization algorithm, and rolling bearing faults were identified by the envelope spectrum analysis of the reconstructed signals in the time domain. Numerical simulation and experimental results demonstrate that the proposed method has good anti-noise performance and may effectively extract weak fault features of rolling bearings under strong noise interference.

Key words:  , rolling bearing of wind turbine, fault diagnosis, group sparse, graph regularization

摘要： 风电机组的非平稳运行、嘈杂环境以及强电磁干扰等影响，使得滚动轴承故障脉冲易被强噪声淹没，微弱特征难以准确识别。提出了一种图正则化约束的频域组稀疏模型，在不依赖周期先验的前提下，实现滚动轴承故障特征的有效提取。将振动信号转化成图信号以构造图正则化约束，利用结构化信息指导惩罚力度，提高稀疏重构的准确性；构建图正则化约束的频域组稀疏模型，给出了组内分量收缩阈值的确定方法，并利用近端映射来简化目标函数以优化求解；最后，利用构造的综合评价指标和蛾焰优化算法优化模型参数，通过对重构后时域稀疏信号的包络谱分析识别滚动轴承故障。数值仿真和实验结果表明，所提方法具有良好的抗噪性能，能够有效地提取强噪声干扰下滚动轴承的微弱故障特征。

关键词: 风电机组滚动轴承, 故障诊断, 组稀疏, 图正则化