Research on high-speed train bogie condition monitoring and fault intelligent diagnosis based on deep learning

Abstract

The bogie is the most crucial component of a high-speed train’s (HST) running gear, directly impacting operational safety and passenger comfort. Major failures in bogie components can lead to severe vibration, performance degradation, and even catastrophic accidents such as derailments. To address this, we propose IntelliGraph-EVO, a novel fault diagnosis framework integrating an Intelligent Graph Convolutional Network (IntelliGraph) and the Energy Valley Optimizer (EVO). IntelliGraph captures structural relationships between bogie components through graph convolutions, while EVO optimizes model parameters to enhance diagnostic performance using sensor-derived synthetic vibration data. Two datasets are utilized: Dataset 1 simulates seven operational conditions (normal and six fault states) at varying speeds (80–200 km/h), and Dataset 2 includes 15 conditions (14 single-fault and normal states) at a fixed 200 km/h. Raw sensor data is pre-processed via min-max normalization, and Fast Fourier Transform (FFT) extracts frequency-domain features. Experimental results demonstrate that IntelliGraph-EVO outperforms state-of-the-art methods like ICEEMDAN + 1-D CNN, achieving precision (0.995), recall (0.994), F1-score (0.994), and accuracy (0.994) for Dataset 1, and accuracy (0.993) for Dataset 2. The model’s ability to diagnose single and mixed faults under dynamic operational conditions highlights its robustness and generalization. This work underscores the potential of deep learning for real-time condition monitoring in HSTs, enabling early fault prediction and proactive maintenance to enhance safety and operational efficiency.