Multimodal data fusion-driven fault diagnosis of aero-engine bearings using a pseudo-Siamese neural network

Abstract To address the challenges of aero-engine inter-shaft bearing faults under complex operating conditions and binary classification sample imbalance, this study proposes a novel fault diagnosis method based on multi-modal data fusion. A three-channel pseudo-Siamese neural network is constructed, where time-domain vibration features are extracted via a simplified one-dimensional variational kernel CNN and a one-dimensional physics-guided CNN, while time-frequency features are derived from Gramian angular summation field images using a two-dimensional CNN for subsequent feature fusion. Correlation analysis confirms the model’s cross-modal complementarity, effectively mitigating information redundancy and conflict. To enhance training efficiency and classification performance, the crested porcupine optimizer (CPO) integrated with a gated recurrent unit is employed for hyperparameter tuning. Evaluated on the Harbin Institute of Technology and Case Western Reserve University bearing datasets with various imbalance ratios, the proposed method achieves a classification accuracy of over 98.5%, outperforming existing methods. Finally, ablation and anti-noise experiments demonstrate that the proposed multi-modal architecture combined with the CPO optimization strategy provide a robust solution for imbalanced fault diagnosis under complex operating conditions.