A Multi-Train Brake System Fault Diagnosis Method Based on An Improved PJR-GAFN Deep Learning Model

With the increasing complexity of modern rail transit, ensuring the operational safety and reliability of multi-train braking systems has become increasingly important. However, traditional fault diagnosis methods often struggle to effectively handle the inherent nonlinear coupling, noise interference, and cross-group correlation in multi-train braking data. To overcome these limitations, this paper proposes an improved Parallel Joint Residual-Graph Attention Fusion Network (PJR-GAFN) for intelligent fault diagnosis of multi-train braking systems. This model combines residual feature extraction with graph attention fusion to jointly capture local component dynamics and global system dependencies. Specifically, the parallel residual learning module enhances feature representation and gradient stability, while the adaptive graph attention fusion module models the multi-source relationships between different braking units. Experimental results on a real-world train braking data-set demonstrate that the improved PJR-GAFN achieves superior diagnostic accuracy, robustness, and generalization compared to state-of-the-art deep learning baselines. Furthermore, ablation studies confirm the effectiveness of each model component in improving fault identification under noisy and cross-correlated conditions. This study provides a novel, interpretable, and computationally efficient approach for intelligent fault diagnosis and predictive maintenance of modern railway systems.