Pipelines are an important type of transportation and are often buried underground, making regular maintenance and inspection challenging, especially for pipelines without network maps. Ultrasonic testing (UT) is a commonly used non-destructive method to assess surface or subsurface defects in pipeline. In this study, we proposed to correlate the UT signals with the geometric and spatial features to reconstruct the network of pipeline. The feasibility of the proposed method has been discussed numerically. Two key geometric features (pipe length and connections) were investigated to explore the correlation between ultrasonic guided wave features and different length and node conditions. This study used principal component analysis to select the characteristics, and integrated the backpropagation neural networks (BPNN) and radial basis function neural networks (RBFNN) to process the signals to establish the relationship between UT signal and spatial features. The results of the study show that BPNN performs better in pipeline length and connection type recognition, with an average coefficient of determination of 0.96 for recognizing the length and an average correct rate of 91.9% for recognizing the connection type. A comprehensive comparison of the two intelligent algorithms reveals that the BPNN performs well in improving the prediction of pipeline complexity, which significantly enhances the detection of geometric and spatial features of pipelines.
|