A novel text-style sequential modeling method for ultrasonic rail flaw detection

Integrity of rails is the foundation of safe rail transportation. It is critical to detect internal rail flaws in time, and one popular solution to this issue is ultrasonic techniques. On the other hand, long short-term memory (LSTM) has been proven in text classification to which we think the ultrasonic rail flaw detection can be quite similar. In this context, this paper proposes a novel text-style sequential modeling method for ultrasonic rail flaw data and a LSTM-based deep learning model for rail flaw detection. Comparative experiments proved the feasibility and remarkable computational efficiency of the proposed modeling method and model.