Temperature-adaptive artificial neural network model for enhanced fault detection in railway tracks using acoustic emission sensor data

Maintaining railway infrastructure integrity is essential for global trade and transportation, as heavy traffic and adverse environments heighten the risk of creep and fatigue-related failures. Acoustic Emission (AE) techniques, widely used in non-destructive testing (NDT), enable real-time health monitoring by detecting stress waves generated during fault nucleation. However, temperature variations significantly influence the wave propagation velocity, affecting the accuracy of fault localisation by altering the Time of Arrival (TOA). This study addresses these challenges by developing an Artificial Neural Network (ANN) model that incorporates dynamic temperature variations to enhance fault localisation accuracy. The ANN is trained using laboratory data at standard temperatures and tested on field data collected across a wide temperature range (28°C to 65°C). The initial results revealed discrepancies in fault prediction accuracy, prompting model refinement by including temperature as a critical input alongside conventional AE parameters. The enhanced model demonstrates significant improvements in fault localisation accuracy, emphasising the importance of accounting for temperature effects. This research contributes to the advancement of AI-driven solutions for reliable real-time fault detection, promoting safer and more efficient railway operations under diverse environmental conditions.