Multisource Fusion Self-Calibrated U-Shaped Network for Hydraulic System Fault Diagnosis

Reliable condition awareness is vital for modern aircraft and industrial equipment, yet fault diagnosis of hydraulic systems remains challenging due to hidden failure modes, nonlinear dynamics, and multi-source heterogeneity. To address this, an end-to-end intelligent diagnosis framework that integrates multi-source fusion self-calibrated U-shaped network (MFSCU-Net)is proposed. Displacement, pressure, and flow signals are fused across data and feature levels, while a U-shaped encoder-decoder with Self-Calibrated convolutions enables multi-scale representation, adaptive recalibration, and deep feature interaction. This design enhances weak fault signatures and improves sensitivity to incipient faults. Experiments on electrohydrostatic actuator and electrohydraulic proportional servo valve datasets demonstrate that MFSCU-Net achieves superior accuracy and robustness over benchmark methods.