Resumen de Pioneering Precision: Unsupervised K-means Clusteringand Attention-CNN-1D for Railway Corrugation Estimation

Masoud Haghbin, Juan Chiachío Ruano, Sergio Muñoz Abad, José Luis Escalona Franco, Antonio J. Guillén, Adolfo Crespo Márquez, Sergio Cantero Chinchilla

• This study introduces a deep learning approach for estimating corrugation in railwaysystems by Attention-based One-Dimensional Convolutional Neural Networks (Attention-CNN-1D) coupled with Unsupervised K-means clustering using acceleration in longitudinaland vertical directions. The model’s performance is examined in a 1:10 scale vehicle runningat two forward velocities: 0.5 and 1.00 m/s. In addition, the newly developed model’s abilityto train with one velocity and test at another was analyzed. The findings indicated that themodel performed accurately across different velocities and combinations. The study shows thatunsupervised K-means- Attention-CNN-1D can potentially improve corrugation estimation. Thiscan lead to more reliable and efficient maintenance and repair of railways.