Real-time rail defect detection using a three-phase eddy current sensor array and lightweight transformer for sustainable transportation

Railway transportation offers significant advantages for low-carbon mobility, provided that infrastructure integrity can be efficiently maintained with minimal service interruptions. This study presents a real-time rail defect detection system combining a three-phase excitation eddy current sensor array with a lightweight Transformer-based deep learning model. The sensor features a hexagonal array arrangement that achieves strong background signal cancellation and high lift-off tolerance, enabling reliable operation under dynamic conditions. Real-time conductivity mapping is performed through a rapid linear back-projection algorithm, and defect classification is accomplished using a Transformer model trained on COMSOL-simulated datasets. Experimental validation demonstrates a classification accuracy exceeding 95%, reliable identification of defects spaced as closely as 5 mm apart, and robustness across lift-off variations up to 10 mm. The system achieves an inference time of less than 0.02 seconds, supporting high-speed rail inspection at velocities up to 10.8 km/h without compromising detection performance. By facilitating continuous rail monitoring and reducing inspection-related energy consumption, the proposed system enhances the operational sustainability of railway infrastructure and contributes to broader low-carbon transportation initiatives. Future work will focus on expanding detection capabilities to subsurface and complex defects, further strengthening the system’s applicability for sustainable rail operation.