TY - JOUR
T1 - VA-TFSC-Focusgram
T2 - a multi-candidate collaborative framework for industrial gearboxes weak fault diagnosis based on vibration-acoustic cross-modal guidance
AU - Li, Laixing
AU - Ying, Wanming
AU - Wang, Teng
AU - Ni, Qing
AU - Noman, Khandaker
AU - Li, Yongbo
N1 - Publisher Copyright:
© 2026 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies.
PY - 2026/9/1
Y1 - 2026/9/1
N2 - Early detection of weak faults is vital for the reliable maintenance of rotating machinery. However, diagnostic methods based on single-modal vibration or acoustic signals are susceptible to interference from background noise, and blind demodulation indexes face the challenge of distinguishing between periodic impulses, random impulses, and meshing harmonics. To address this issue, this paper proposes a multi-candidate collaborative search method driven by cross-modal physical guidance and decision fusion. A novel blind metric, termed the improved enhanced cyclostationary sparsity index (IECSI), is developed as the key criterion for optimal demodulation frequency band (ODFB) selection. It combines the sparsity of the Gini index with the quantification of cyclostationarity via autocorrelation spectrum entropy. By further introducing an impulse penalty operator, IECSI effectively suppresses interference from meshing harmonics and random noise. This framework first achieves reliable coarse localization of ODFB through vibration-acoustic time–frequency spectral correlation (VA-TFSC). Building on this physical anchoring, an adaptive ternary tree search strategy is developed to construct Focusgram. Finally, a Top-P multi-candidate extraction strategy, together with cross-modal joint score and adaptive arbitration, is introduced to identify physically consistent frequency pairs and intelligently select high-quality diagnostic channels. Simulation and experimental results demonstrate that the VA-TFSC-Focusgram significantly enhances weak fault feature extraction robustness, maintaining a stable diagnostic boundary even under extreme low-SNR conditions.
AB - Early detection of weak faults is vital for the reliable maintenance of rotating machinery. However, diagnostic methods based on single-modal vibration or acoustic signals are susceptible to interference from background noise, and blind demodulation indexes face the challenge of distinguishing between periodic impulses, random impulses, and meshing harmonics. To address this issue, this paper proposes a multi-candidate collaborative search method driven by cross-modal physical guidance and decision fusion. A novel blind metric, termed the improved enhanced cyclostationary sparsity index (IECSI), is developed as the key criterion for optimal demodulation frequency band (ODFB) selection. It combines the sparsity of the Gini index with the quantification of cyclostationarity via autocorrelation spectrum entropy. By further introducing an impulse penalty operator, IECSI effectively suppresses interference from meshing harmonics and random noise. This framework first achieves reliable coarse localization of ODFB through vibration-acoustic time–frequency spectral correlation (VA-TFSC). Building on this physical anchoring, an adaptive ternary tree search strategy is developed to construct Focusgram. Finally, a Top-P multi-candidate extraction strategy, together with cross-modal joint score and adaptive arbitration, is introduced to identify physically consistent frequency pairs and intelligently select high-quality diagnostic channels. Simulation and experimental results demonstrate that the VA-TFSC-Focusgram significantly enhances weak fault feature extraction robustness, maintaining a stable diagnostic boundary even under extreme low-SNR conditions.
KW - Early detection
KW - Focusgram
KW - Improved enhanced cyclostationary sparsity index
KW - Time-frequency spectral correlation
UR - https://www.scopus.com/pages/publications/105046557024
U2 - 10.1016/j.ymssp.2026.114784
DO - 10.1016/j.ymssp.2026.114784
M3 - 文章
AN - SCOPUS:105046557024
SN - 0888-3270
VL - 259
JO - Mechanical Systems and Signal Processing
JF - Mechanical Systems and Signal Processing
M1 - 114784
ER -