Development of Smart Real-time Fault Detection Approach in Railway Track Deploying a Single Acoustic Emission Sensor Data

Railways, integral to global trade and transportation, face infrastructure vulnerabilities from heavy traffic and challenging environments. Timely fault monitoring is pivotal for effective risk mitigation. Acoustic emission (AE) techniques, particularly in non-destructive testing (NDT), provide real-time health monitoring for railway track. However, traditional AE methods, especially signal processing, prove complex and time-intensive for fault localization. Recent studies explore the potential of Artificial Intelligence (AI), specifically neural networks, for fault detection, yet lack a systematic approach to parameter weighting.

This study bridges this gap by systematically evaluating weights assigned to the AE parameters to enhance fault detection accuracy. Laboratory and field-tested AE sensor data from railway track assess the significance of parameters such as Amplitude, Peak Frequency, Counts, RMS, Energy, Rise Time, Duration. As carriers of crucial fault information within the AE signal. A novel methodology introduces distinct weights to individual AE parameters based on their importance, refining the AI model's focus on critical attributes. Extensive laboratory experiments simulate damage in various rail sections, assessing the AI model's fault localization efficiency, robustness, and accuracy using an Artificial Neural Network (ANN) model.

The ANN model demonstrates superiority in precise fault localization, affirming its efficacy. The proposed methodology presented herein, validated through rigorous assessments in both controlled laboratory environments and real-world field conditions, stands as a noteworthy advancement in the realm of smart fault detection in real-time.