Pressure-Only Diagnosis of External Gear Pumps via Ground-Test Modality Augmentation and Physics-Guided Feature Enhancement

In aviation and industrial applications, onboard fault diagnosis of external gear pumps is often restricted to pressure pulsation signals due to hardware limitations, which may restrict fault detection capacity, whereas laboratory ground tests can collect multimodal data for enhanced analysis. To address this gap, we propose a two-level enhancement framework. At the data-driven level, vibration and pressure signals from ground tests are fused to learn shared latent representations, enabling complementary diagnostic cues. At the physics-guided level, fluid dynamics?based modeling extracts interpretable fault features from pressure signals, embedding domain-specific priors. This dual enhancement distills both data-driven and physics-based knowledge into the pressure modality, allowing the model to achieve high diagnostic accuracy under single-modality conditions. Experiments on an external gear pump confirm that vibration features strengthen sensitivity to structural faults, while physics guidance enhances interpretability by anchoring features to frequency and amplitude variations. The model was further deployed on a hardware platform to validate real-time performance. The code and demo video after model deployment are available at https://github.com/ts560/Net.