基于移不变字典学习和稀疏编码的滚动轴承故障识别算法

针对现有旋转机械故障识别算法过度依赖专家先验知识的问题，提出了一种基于移不变字典学习和稀疏编码（SIDL-SC）的自适应故障识别算法。首先，将不同故障状态下的振动信号进行分段和平滑预处理以降低数据处理复杂度，接着将加入自适应惩罚因子的移不变字典学习算法用于提取不同故障状态下的移不变基函数；然后，利用高效的特征符号搜索算法求解待识别信号在不同基函数下的稀疏系数以实现对待识别信号的重构；最后，以重构残差作为对该信号故障状态识别的判断依据。滚动轴承振动数据库和实测航空发动机振动信号的实验结果表明，该算法相较于现有算法具有更高的故障识别准确率，在实际中具有较强的可行性。

Fault recognition algorithm for rolling bearings based on shift invariant dictionary learning and sparse coding

According to current algorithms for rotating machines largely depending on expert prior knowledge, the paper proposed an adaptive fault recognition algorithm based on shift invariant dictionary learning and sparse coding. Firstly, it segmented and smoothed vibration signals to decrease the complexity. Then, it used shift invariant dictionary learning with adaptive penalty factor to learn shift invariant bases in different fault states. After that, it used an efficient sparse coefficient solver called Feature Sign Search for reconstructing signal to be recognized. Lastly, residual was an evidence to determining fault state the signal belonging to. In the experiments of rolling bearing datasets and vibration signals of real aero-engine demonstrate its higher accuracy than up-to-date algorithms and feasibility for practical applications.