A Gearbox Vibration Signal Compressed Sensing Method Based on the Improved GLOW Flow Model

In response to the complex characteristics of gearbox vibration signals, including high frequency, high dimensionality, non-stationarity, non-linearity, and noise interference, this paper proposes a data processing method based on improved compressed sensing.First, the K-means Singular Value Decomposition (K-SVD) dictionary is used for sparse representation, ensuring good sparsity in the frequency domain.Next, a random convolution kernel measurement matrix is employed in place of the traditional Gaussian random matrix, satisfying the equidistant constraint while enhancing both computational and hardware implementation efficiency.Finally, the Generative Flow (GLOW) model is introduced, incorporating the measurement matrix, dictionary matrix, and sparse coefficient matrix into a unified optimization framework for joint solving.Through reversible mapping and probabilistic distribution modeling, the method effectively addresses noise interference and the challenges posed by complex signal distributions.Experimental results show that, compared with traditional compressed sensing methods, the proposed method offers superior signal reconstruction quality and better noise robustness.