This paper describes a novel approach to determine the wall thinning of piping utilizing the deep neural network (DNN) with the frequency variation of guided wave reflection coefficients as vector data to the input layer of the DNN. First in this paper, it is shown that the frequency variation of the guided wave reflection coefficients has important information about defect shape [1], so we used those in the DNN. It is also shown as a specific feature of the method that the enough training data (over 210000) for the DNN, which are consist of many pairs of the reflection coefficients and its correspondent wall thinning, are gathered fast (consuming several minutes or so) by calculating a mathematical model [1]. The characteristics of the mathematical model was described and those were confirmed to be approximately similar characteristics to the FEM calculation results. Frequency variation of 25 kHz to 70 kHz (1 kHz step) was used as 10-dimensional vector data to the input layer of the DNN. Three intermediate layers, those all are having 1000 units, were installed. The output layer was installed with the softmax function for the multinominal classification of wall thinnings into 0.5 mm steps. The evaluations of the DNN trained with the data calculated by the mathematical model were carried out with the two different test sets. One is the set of the reflection coefficients obtained from 21 different concave artificial defects in 50A schedule40 pipes (three concave curvatures R=50, 100, and 200 mm, and seven different wall thinnings d=0.5, 1.0, 1.5, 2.0, 2.5, 3.0, and 3.5 mm). In the set, it was confirmed that more than 90% (19/21) correct answer rate was obtained in ± 0.5 mm error. The other is the set of the reflection coefficients obtained by FEM calculations from two actual wall thinning distributions in 80A shedule40 pipes from a chemical plant. The estimation results were coincided fairly well with the actual wall thinnings. All the classification results are shown and discussed.&nbsp;Reference:[1] Hideo Nishino, “An investigation of reflection coefficients of the T(0,1) mode guided waves at axisymmetric defects and inverse problem analyses for estimations of defects shapes”, Materials Transaction 56 (2015) 120-128.

Multinominal Classification of Wall Thinning of Piping Using a Deep Neural Network Based on the Frequency Variation of Guided Wave Reflection Coefficients at Defect

Paper Proceedings: https://asmedigitalcollection.asme.org/qnde/QNDE2021/volume/85529

ASME 2021 48th Annual Review of Progress in Quantitative Nondestructive Evaluation (QNDE2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.