Rotating machinery fault diagnosis based on sparse assisted invariant signal model

In rotating machinery operating under complex, variable, and high-noise conditions, early-stage fault signals are frequently masked by low-frequency trends, rotational frequency modulations, and random noise. These interferences significantly impair the performance of conventional fault feature extraction methods. To address these limitations, this study proposes a fault feature extraction framework based on the sparse-assisted shift-invariant signal model (SAISM). This approach integrates trend suppression and shift-invariant sparse modeling to enhance structural features and precisely extract weak impact signals. A sparse-assisted smoothing operator is first applied to preprocess the raw signal, effectively suppressing low-frequency trends and broadband noise and thereby improving the signal-to-noise ratio for subsequent impact separation. The preprocessed signal is then segmented, after which shift-invariant sparse coding (SISC) is utilized to construct an adaptive dictionary capturing periodic impact responses, facilitating the detection of weak fault features. The reconstructed sparse signal is subsequently analyzed using envelope spectrum techniques to identify fault characteristic frequencies. Experimental results validate the effectiveness of SAISM in highly noisy environments. Comparative tests on the XJTU-SY bearing degradation dataset demonstrate its superior robustness and accuracy relative to traditional methods such as wavelet packet decomposition (WPD), underscoring its potential for practical engineering applications.