Wheel flat detection using long short-term memory and transformer models with a 1:10 scale railway test rig

Abstract

In railway systems, the detection of wheel flats is essential for ensuring safety and reducing maintenance costs. This study compares the performance of Long Short-Term Memory and Transformer models in detecting wheel flats using data from a 1:10 scale railway test rig. The findings indicate that the Transformer model significantly outperforms the Long Short-Term Memory model, especially when feature-level sensor fusion is employed, achieving an average error as low as 0.0069 mm with percentage of error at 5.30%, minimizing the maximum error to 0.0985 mm. The study emphasizes the potential of Transformer models in railway diagnostics, particularly for applications requiring high accuracy and reliability. The insights gained from this research have practical implications for improving the precision of wheel flat detection in real-world railway operations, enhancing both safety and efficiency.