Ultrasonic guided wave imaging with deep learning: Applications in corrosion mapping

In this paper, a rapid guided wave imaging method based on convolutional neural network (CNN) is proposed to quantitatively evaluate the corrosion damage. The method contains offline training and online imaging. The purpose of offline training is to establish the relationship between the detection signals and the velocity map based on forward modeling data. In the step of online imaging, the velocity map can be predicted in real-time with the detection signals fed into the trained model. Then, the remaining thickness of corroded structures can be estimated according to the dispersion curves of a specific guided wave mode. Numerical results indicate that the average correlation coefficients of the optimal model are respectively 0.9493, 0.9273, and 0.9262 in training, validation, and testing. The success rate of applying the optimal model to the testing set is 82.73% if the correlation coefficient greater than or equal to 0.9 is used as the criterion of successful corrosion imaging, which proves the excellent imaging performance. Furthermore, the imaging speed is verified and the damage reconstruction of 4000 samples is done within 3 s. The imaging method also can be used to detect the position of small corrosion damage. For a noise contaminated dataset, the size and location can be accurately predicted, albeit damage sizing is rather challenging. Moreover, experiments have been carried out and the correlation coefficient between the true velocity map and the imaging results is 0.9109, which proves the imaging method can be applied in practice.

Automated summary

In this paper, a rapid guided wave imaging method based on convolutional neural network (CNN) is proposed for quantitative evaluation of corrosion damage. The method involves offline training and online imaging, and has shown excellent imaging performance and high success rate in numerical experiments.