Spatio-Temporal Joint Network for Coupler Anomaly Detection Under Complex Working Conditions Utilizing Multi-Source...

Area of Science:

• Mechanical Engineering
• Data Science
• Artificial Intelligence

Background:

• Extracting spatio-temporal features from high-dimensional sensor data under fluctuating loads is challenging.
• Intricate mechanical coupling in systems like heavy-haul trains complicates anomaly detection.
• Existing methods struggle with noise and accurately capturing complex system dynamics.

Purpose of the Study:

• To develop a robust anomaly detection framework for critical load-bearing components in heavy-haul trains.
• To improve the accuracy and reliability of condition monitoring systems.
• To provide a data-driven analytical framework for early identification of equipment damage.

Main Methods:

• Utilizing Normalized Mutual Information (NMI) to quantify sensor coupling and filter noise.
• Employing Spatio-Temporal Graph Neural Networks (STGNN) with Graph Convolutional Networks (GCN) and Gated Recurrent Units (GRU) for feature extraction.
• Implementing a moving average algorithm and the 3σ criterion for anomaly detection and early warning.

Main Results:

• The proposed framework demonstrated superior fitting performance and noise robustness compared to CNN and LSTM models.
• The method effectively reduced false alarm rates during normal operation.
• Real-world case studies successfully identified local damage and fatigue crack initiation sites with high spatial consistency.

Conclusions:

• The NMI-STGNN framework offers a viable data-driven solution for condition monitoring and anomaly identification in heavy-haul trains.
• The approach enhances the ability to detect subtle changes in spatial linkage features indicative of damage.
• This study contributes a robust method for ensuring the safety and reliability of critical infrastructure.