High-Resolution Thickness Mapping with Deepfit and Lamb Guided Waves

Accurately predicting the remaining wall thickness of metal structures like plates, pipes and pressure vessels is of significant importance to the petrochemical industry. However, traditional ultrasonic probing techniques demand point by point scan of the target structures, which costs enormous time and money. In this paper, we present a robust guided wave tomography algorithm, DeepFIT. The algorithm adopts a neural network to approximate the execution of descent direction matrix in fast inversion tomography (FIT). To achieve robust imaging, signal components and corresponding phase velocity maps of A0 mode Lamb guided waves are input into DeepFIT for training. This technique guarantees that the inversion process can be significantly accelerated, circumventing the enormous computational burden caused by Hessian and Jacobian matrix calculation in full waveform inversion (FWI). The proposed method builds the foundation for fast and robust quantitative industrial inspection with Lamb guided waves.