航天器传动部件振动数据稀疏多分类智能故障诊断

针对航天器传动链机械零部件发生故障时训练数据稀缺以及信号的强非平稳\非线性特点,提出了动态数据驱动的子空间稀疏多分类智能故障诊断算法。首先对采集的单一传感器动态信号进行分形小波变换,生成常规和非常规二进小波尺度。然后对生成的各子空间提出基于故障能量指标的稀疏多分类器,以监测部件的故障特征频率为搜索目标,在子空间的包络解调谱上计算特征频率及其倍频附近的能量峰值,导出特定故障模式的稀疏评价指标,以各种故障模式的最大值识别和判定故障类型。所提出的算法完全实现了无人工监督的智能故障诊断。最后以航天器轴承故障诊断为例验证了该算法的有效性。所研究算法的泛化能力较强,技术路线同样适用于其他航天器传动部件的在线监测与故障智能预警。

Sparsity assisted intelligent multi-classifier for fault diagnosis using vibration data of spacecraft transmission chain

In view of limited training samples and strong non-stationarity and nonlinearity for the faults of the transmission chain within spacecraft, we propose a novel intelligent sparsity assisted multi-classifier based diagnostic algorithm of using the dynamic data. Firstly, the dynamic data, from single sensor acquisitions, are decomposed via the fractal wavelet, to generate the conventional dyadic and non-conventional wavelet subscales. The fractal wavelet is constructed based on the nearly translation-invariant complex wavelet bases and the transform of the classical dyadic wavelet is used for extracting the transition-band dynamic features. On the other hand, the decomposed wavelet subscales are further processed by an intelligent indicator of the fault energy ratio based sparsity multi-classifier. This method can describe fault severity based on the characteristic frequencies of the component faults and generate quantitative sparsity indicators. The proposed method can achieve the goal of the automatic fault diagnosis without artificial interferences. A case study based on the spacecraft bearing is employed to validate the effectiveness of the proposed algorithm. The processing results successfully indicate the types of the actual faults in the bearing.