A novel flaw detection approach in carbon steel pipes through ultrasonic guided waves and optimized transformer neural networks

MF Flores Cuenca, AK Malikov, J Kim, Y Cho… - Journal of Mechanical …, 2024 - Springer
MF Flores Cuenca, AK Malikov, J Kim, Y Cho, K Jeong
Journal of Mechanical Science and Technology, 2024Springer
In the present study, ultrasonic guided wave testing (UGWT) is used as a nondestructive
inspection technique to identify flaws in carbon steel pipe. The selection of a wave mode in
UGWT is critical. Therefore, the flexural wave mode F (1, 1) has been found to be a suitable
choice. Not only that, but the pipe's length is also subject to variation, and the defect may
manifest at disparate locations along its length. To address this, several studies have been
explored such as convolutional neural networks and recurrent neural networks …
Abstract
In the present study, ultrasonic guided wave testing (UGWT) is used as a nondestructive inspection technique to identify flaws in carbon steel pipe. The selection of a wave mode in UGWT is critical. Therefore, the flexural wave mode F (1, 1) has been found to be a suitable choice. Not only that, but the pipe’s length is also subject to variation, and the defect may manifest at disparate locations along its length. To address this, several studies have been explored such as convolutional neural networks and recurrent neural networks. Nonetheless, a novel technique is proposed in this research, the transformer neural network architecture due to its capability to capture long-range dependencies and temporal relationships. The accuracy of the chose deep learning architecture was evaluated, and suitable hyperparameters were selected to optimize the model. As a result, the transformer architecture accurately predicts the flaw pattern for the vast majority of the test data, attaining an accuracy and F1-score of 99 %.
Springer
以上显示的是最相近的搜索结果。 查看全部搜索结果