Barat V.A., Marchenkov A.Yu., Karpova M.V., Bardakov V.V., Lepsheev E.A., Ushanov S.V., Elizarov S.V. «Application of artificial neural networks for detection of defects in dissimilar welded joints by acoustic emission method» Контроль. Диагностика, 27, № 12, с. 4-13 (2024)

The paper considers the possibility of using artificial neural networks to detect hits in acoustic emission (AE) testing. A distinctive feature of the proposed method is that the training set of the neural network is formed using a complex technique based on the application of modeling technology, on the one hand, and on calibration measurements carried out in the field, on the other. In this paper, process pipelines with dissimilar welded joints were considered as a test structure. AE signals were modeled using a hybrid method: the signal waveform was determined based on a finite element model, and the AE hits amplitudes were determined on the basis of a physical experiment on cyclic stretching of samples of dissimilar welded joints. Acoustic signals measured on the process pipelines bodies in the field condition were used as noise. A multilayer perceptron was used to classify the data, the architecture of which was selected based on the minimization of the classification error. Keywords: dissimilar welded joints, acoustic emission, diffusion layers, neural networks in acoustic emission, classification of acoustic emission signals, waveguide modeling.

Barat V.A., Marchenkov A.Yu., Karpova M.V., Bardakov V.V., Lepsheev E.A., Ushanov S.V., Elizarov S.V. «Application of artificial neural networks for detection of defects in dissimilar welded joints by acoustic emission method» Контроль. Диагностика, 27, № 12, с. 4-13 (2024)

The paper considers the possibility of using artificial neural networks to detect hits in acoustic emission (AE) testing. A distinctive feature of the proposed method is that the training set of the neural network is formed using a complex technique based on the application of modeling technology, on the one hand, and on calibration measurements carried out in the field, on the other. In this paper, process pipelines with dissimilar welded joints were considered as a test structure. AE signals were modeled using a hybrid method: the signal waveform was determined based on a finite element model, and the AE hits amplitudes were determined on the basis of a physical experiment on cyclic stretching of samples of dissimilar welded joints. Acoustic signals measured on the process pipelines bodies in the field condition were used as noise. A multilayer perceptron was used to classify the data, the architecture of which was selected based on the minimization of the classification error. Keywords: dissimilar welded joints, acoustic emission, diffusion layers, neural networks in acoustic emission, classification of acoustic emission signals, waveguide modeling.