Gearbox compound fault diagnosis using CEEMDAN feature extraction and a dual-attention multi-scale BiLSTM model

As a core component of mechanical transmission systems, the gearbox's operating state directly determines equipment reliability and industrial production safety. In actual working conditions, a single fault can easily evolve into a complex fault mode with multiple coupled faults. Traditional diagnostic methods face challenges such as insufficient feature extraction and low fault mode discrimination. To address this issue, an intelligent diagnostic model is proposed that integrates adaptive noise complete set empirical mode decomposition (CEEMDAN) feature extraction, multi-scale convolution, and a dual attention mechanism. First, CEEMDAN is used to decompose the vibration signal at multiple scales. After effective IMF filtering, time-domain, frequency-domain, fault-specific, and coupled interactive features are extracted to form a multi-dimensional feature set. Then, adaptive principal component analysis (PCA) is used to reduce the dimensionality to obtain a low-redundancy feature set. Subsequently, a diagnostic model containing multi-scale convolution, a bidirectional long short-term memory network (BiLSTM), and dual attention branches is constructed, and an improved loss function is combined to enhance the ability to distinguish complex fault features. Experimental results based on the Beijing Jiaotong University bogie gearbox bench dataset verify the effectiveness and robustness of the proposed method under complex fault modes, providing a reliable technical solution for gearbox fault diagnosis in industrial scenarios.