Fast Prediction for Dispersion Curves in Pipes Based on Neural Network

Ultrasonic guided waves, known for their high sensitivity and long detection distance, are widely studied in non-destructive testing and structural health monitoring. As the theoretical foundation of ultrasonic guided wave testing, calculating dispersion curves accurately and fast is critical, enabling the study of dispersion phenomena and gaining a deeper understanding of propagation mechanisms. However, existing methods, like the global matrix method and the semi-analytical finite element method often involve high computational costs and time-consuming processes. In this study, deep learning is applied for the first time to predict dispersion curves in pipes to overcome these challenges. Results show that this method can accurately predict both single-mode and multi-mode dispersion curves in pipes, with the mean square error less than 0.1%. In addition, all predictions can be completed in less than three seconds, providing near real-time, highly accurate, and fully automated prediction that is significantly faster - almost thirty times faster - than traditional methods.