A Reliability-Oriented Multi-View Feature Fusion Framework for Imbalanced Fault Diagnosis of Electrical Submersible Pumps under Device-Aware Validation

The Electrical Submersible Pump (ESP) is a critical artificial lift system in oil and gas production, and reliable condition monitoring and fault diagnosis are essential for ensuring safe equipment operation and reducing maintenance costs. However, ESP fault diagnosis in practical industrial environments still faces challenges, including class imbalance, insufficient feature representation, and data distribution differences among devices. Conventional random partition-based evaluation approaches cannot accurately reflect model performance under device distribution variation conditions.To address these issues, this study establishes a reliable diagnostic framework for ESP fault diagnosis. By integrating expert features with clear mechanical fault mechanism information and latent spectral representations, the proposed framework achieves complementary multi-view feature representation. Meanwhile, a leakage-free model development procedure is established, and a combination of random stratified validation and Hybrid Device-Aware validation strategies is employed to systematically evaluate model stability under varying device conditions.Experimental results on the publicly available ESPset dataset demonstrate that the proposed method achieves Macro F1-scores of 0.8856 and 0.7648 in Scenario I and Scenario II, respectively. Ablation studies and multi-model comparisons further verify the effectiveness of the feature representation, class balancing strategy, and model optimization procedure.The results indicate that the proposed method can provide a reference for the development and reliable evaluation of industrial fault diagnosis models under small-sample, imbalanced, and device distribution variation conditions.