Abstract：Ultrasonic guided wave technology has played a significant role in the field of nondestructive testing due to its advantages of high propagation efficiency and low energy consumption. At present, the existing methods for structural defect detection and quantitative reconstruction of defects by ultrasonic guided waves are mainly derived from the guided wave scattering theory. However, taking into account the high complexity in guided wave scattering problems, assumptions such as Born approximation used to derive theoretical solutions lead to poor quality of the reconstructed results. Other methods, for example, optimizing iteration, improve the accuracy of reconstruction, but the time cost in the process of detection has remarkably increased. To address these issues, a novel approach to quantitative reconstruction of defects based on the integration of convolutional neural network with guided wave scattering theory has been proposed in this paper. The neural network developed by this deep learning-assisted method has the ability to quantitatively predict the reconstruction of defects, reduce the theoretical model error and eliminate the impact of noise pollution in the process of inspection on the accuracy of results. To demonstrate the advantage of the developed method for defect reconstruction, the thinning defect reconstructions in plate have been examined. Results show that this approach has high levels of efficiency and accuracy for reconstruction of defects in structures. Especially, for the reconstruction of the rectangle defect, the result by the proposed method is nearly 200% more accurate than the solution by the method of wavenumber-space transform. For the signals polluted with Gaussian noise, i.e., 15 db, the proposed method can improve the accuracy of reconstruction of defects by 71% as compared with the quality of results by the tradional method of wavenumber-space transform. In practical applications, the integration of theoretical reconstruction models with the neural network technique can provide a useful insight into the high-precision reconstruction of defects in the field of non-destruction testing.
2021, Vol. 42 
