基于GST的变速机械故障信号稀疏特征提取方法

为提取强噪声背景下的变速旋转机械设备的冲击故障特征，提出了一种基于广义S变换的稀疏特征提取方法.首先，通过多分辨率广义S变换（multiresolution generalized S-transform，MGST）搜索每次迭代过程中的最佳原子，多分辨率广义S变换可以得到信号不同尺度下的归一化时频谱，并从中找出能量最大值及其所对应的时频因子，根据故障冗余字典的构建模型可得到冲击成分的最佳匹配原子.其次，结合正交匹配追踪算法（orthogonal matching pursuit，OMP），计算出信号在原子集合下的投影，由于采用了基于多分辨率广义S变换的原子搜索策略，大幅度提高了OMP的分解效率.最后，根据稀疏表示中第一个冲击信号的出现时刻，可依次计算出冲击信号在变速情况下的出现时刻理论值，通过与实测值的比较，实现变速机械的故障诊断.仿真和实例分析结果表明，该方法比传统OMP方法和广义S变换具有更高的计算效率和定位精度. 

Sparse Feature Extraction for Variable Speed Machinery Based on Sparse Decomposition Combined GST

In order to extract fault impulse feature of variable speed machinery from strong background noise, a sparse feature extraction method based on sparse decomposition combined generalized S transform (GST) was proposed in this paper. Firstly, multi-resolution generalized S transform (MGST) was used to pursuit the optimal atom in each iteration, to get normalized time-frequency spectrums with different scales, and to find the maximum energy and corresponding time-frequency factors to build an optimal atom. Then, an orthogonal matching pursuit (OMP) was used to decompose the signal into several optimal atoms, and the efficiency of atoms pursuit was improved with MGST. Finally, the theoretical locations of impulses were calculated according to the location of first impulse in the sparse representation signal, and the fault was diagnosed through the comparison of theoretical and measured locations. The results of simulation and experiment validate the performances of the proposed method, being better than traditional GST method and OMP method in precision and decomposition speed.