<p>Repetitive impulsive mode is a signature of rotating machinery with localized defect. Impulsive mode decomposition (IMD) was customized to extract this type mode in recent. However, it suffers from limitations of the low computational efficiency and local optima induced by the random initialization of particles, and the insufficient capability to distinguish different modes under strong noise interference due to not incorporate the spectral information in objective function. To overcome these limitations, this paper proposes a spectral information-enabled IMD (SIEIMD), providing a robust framework for fault-related transient extraction. Guided by the convergence characteristics of the variational model, the spectral information is exploited to rapidly locate the center frequencies of latent modes for the SIEIMD, enabling an informed initialization of particles. A novel time–frequency sparsity measure is constructed as the objective function of SIEIMD, enhancing the characterizing robustness of repetitive transients. One simulated and two experimental cases are conducted to validate the diagnostic performance. The qualitative and quantitative analysis results demonstrate that our SIEIMD significantly outperforms an improved variant of IMD, variational mode decomposition, feature mode decomposition and the Fast Eserogram method in terms of impulsive mode extraction effectiveness and computational efficiency.</p>