The Sedpgram: A Novel Optimal Demodulation Frequency Band Selection Framework for Multitype Gearbox Early Fault Diagnosis

Early fault detection in multitype gearbox remains a critical challenge, as weak fault-induced signals are frequently overwhelmed by strong ambient noise and operational interference. Existing methods are primarily tailored for fault characteristics of a single gearbox type, severely limiting their applicability and robustness in complex industrial gearboxes. While optimal demodulation frequency band (ODFB) analysis is a potent feature extraction technique, its performance is fundamentally constrained by frequency band segmentation strategies and the design of optimization metrics. To address this issue, this article proposes a novel ODFB selection framework on the spectral energy distribution probability (SEDP) gram to robustly extract multitype gearbox fault features. Its framework prioritizes identifying the most informative band center frequency using the SEDP metric—which quantifies the concentration of characteristic energy. Subsequently, a binary tree algorithm is employed to adaptively segment the frequency spectrum, recursively approximating the identified center frequency. Crucially, the demodulation sideband retention (DSR) ratio is used to autonomously determine the optimal segmentation level, ensuring maximal retention of fault information while suppressing noise. This framework enables effective multitype gearbox fault diagnosis solely through preset parameter fine-tuning. Validation using the NWPU multitype gearbox datasets demonstrates that proposed method outperforms traditional ODFB methods in enhancing early fault features.