<p style="text-align: justify">It is widely agreed that machine learning (ML) algorithms have a tremendous potential in NDT and in particular for automated classification or characterization of defects. In this communication, we demonstrate and evaluate this potential in ultrasonic array imaging, with a specific case of interest corresponding to the multi-modal TFM (Total Focusing Method) imaging of crack-like defects. In multi-mode imaging, an accurate characterization of the defect may require extracting information from several images computed with the same collection of signals, each image being associated with a reconstruction mode involving one or more reflections and mode conversions at the specimen interfaces. The selection of the relevant reconstruction modes giving rise to representative images of a crack-like defect is not always a trivial task. Furthermore, in real applications, the quality of &nbsp;TFM images may be degraded due to a lack of knowledge on some influent parameters of the inspection(e.g., actual probe position and orientation, etc.) and the inspected structure as well (e.g., actual geometry of back-wall interface, elastic properties of the material etc.). Within the European project ADVISE, ML algorithms have been trained on a dataset generated with the CIVA simulation platform, and applied to multi-modal&nbsp;TFM images to retrieve the defect characteristics (size, orientation, etc.), taking account of uncertainty on the specimen parameters. In this paper, we focus on the use of deep convolutional neural network architecture specifically designed to employ multi-modal TFM images for performing flaw localization and sizing. The performance of the ML model has been assessed on both numerical and experimental data. Furthermore, the impact of uncertainties on the inversion results is discussed.</p>

Application of Machine Learning to Multi-Modal Tfm Imaging for the Characterization of Crack-Like Defects in the Presence of Uncertainties on the Characteristics in Welds

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.