In-band noise-suppressed bidirectional reweighted cyclostationary analysis for acoustic emission-monitored fault diagnosis

Weak fault-induced transients generated in the early stages of gear drive degradation are frequently buried beneath strong in-band noise caused by complex resonance and modulation phenomena, which poses a major challenge for accurate feature extraction from acoustic emission (AE) signals. To address this issue, this study proposes an In-band Noise-suppressed Bidirectional Reweighted Cyclostationary Analysis (IN-BRCA) method for AE-based fault diagnosis of gear reducers. The proposed framework constructs a spectral coherence matrix as the analytical foundation and introduces two complementary mechanisms. Along the spectral frequency direction, a multi-kernel maximum mean discrepancy measure is employed to adaptively reweight frequency subbands according to their divergence from healthy references, thereby enhancing fault-related spectral contributions. In the cyclic frequency direction, differential residual operations emphasize fault-induced cyclic components that are absent in healthy signals, while a rotational interference suppression weighting mechanism and a harmonic product spectrum strategy jointly amplify weak modulation features and attenuate in-band noise. Extensive experimental verification on planetary gearboxes and harmonic reducers confirms the superior diagnosis performance and robustness of the proposed method compared with several state-of-the-art benchmarks.