混合高斯噪声条件下稀疏表示方法及其在冲击类故障特征提取中的应用

Traditional sparse representation (SR) methods have been widely studied in fault diagnosis field due to their unique advantages in impact feature extraction.However, the traditional SR theory is based on an assumption of Gaussian distribution of interference noise, which makes it difficult to apply to the actual scenario where multiple noise distributions are involved. Regarding the issue above, a new sparse representation method of impact features under mixed Gaussian noise conditionis proposed in this study. Depending on the Bayesian framework of the traditional sparse representation theory and the universal approximation property of the mixed Gaussian distribution, a sparse decomposition model of the mixed Gaussian noiseis established based on the db4 wavelet dictionary, and an optimization algorithm based on Expectation-Maximum (EM) and Alternating Direction Method of Multipliers (ADMM) is derived for model solution. The simulation and experimental results show that the proposed method can effectively extract the weak impact feature under mixed noise interference.