Defects detection for railway catenary system with encoder-decoder architecture

• • The encoder-decoder architecture with residual analysis enables detection of hard points and PSWI in catenary systems, achieving accurate localization without high sampling frequency requirements, thus reducing sensor costs and enabling broader deployment across existing railway networks. • • The quasi-Welch method mitigates edge effects in signal segmentation, improving detection reliability at segment boundaries where conventional methods fail, as evidenced by reduced false positives in field tests. • • Normalization techniques significantly affect defect identification outcomes; improper normalization can degrade detection sensitivity by up to 30%, underscoring the need for careful preprocessing in operational deployments. • • GPS inaccuracies introduce localization errors of up to 5 meters, but the residual-based criterion maintains detection accuracy, ensuring robust performance even with imperfect positioning data.