Numerical investigations for micro-crack evaluation and localization in pipelines using nonlinear ultrasonic guided wave combining deep learning

Abstract

The evaluation and localization of micro-cracks in pipelines were studied by combining nonlinear ultrasonic guided wave and the SEResNet50 network in this paper. The nonlinear effects arising from the interaction of ultrasonic guided wave and micro-cracks in different directions, lengths and locations were investigated using the finite element simulation. In the analysis section, the stacked spectrum map which contains more obvious nonlinear features was introduced to analyze and reveal the impact of these three factors on the spectrum of the signal received by the sensor array. During the learning part, in order to further identify the relationships between the micro-crack features and stacked spectrum map, the SEResNet50 network was provided for the training and conducted for further prediction of the micro-crack with a well-trained model. The results show that the accuracy of the validation set is up to 98.57%, and the generated model also performs well on the test set with an accuracy of 97.22%. The outcome can illustrate that it is feasible for the well-trained SEResNet50 network to identify the complex connections between the spectrum of the sensor array and micro-cracks, enabling the simultaneous evaluation and location of micro-cracks. In summary, the method proposed in this paper combining nonlinear ultrasonic guided wave and deep learning provides a new approach for the detection of micro-cracks in pipelines and will further promote the application of artificial intelligence in non-destructive testing.