For several years, many computer vision applications such as object detection, scene understanding or 3D vision, have benefit from the huge interest of the data processing community&nbsp; for neural networks approaches. Yet, the benefit of such approaches for the optimization of defect detection in materials and structures using non destructive testing (NDT) is still to be investigated. This is the aim of this paper, in which we consider in particular the inspection of composite materials by ultrasonic testing. Hence for this work, we have experimentally built a database of ultrasonic data acquired on composite panels with and without defects. 
&nbsp;&nbsp; &nbsp;A few papers dealing with the use of a neural network for defect detection in ultrasonic data can be found in literature. However, they usually start by a&nbsp; transformation of the 1D signal into a 1D features signal, having a lower size. This transformation is obtained using for instance the well-known wavelet transform or few layers of neural network. This features signal is then put at the entrance of a classification network. 
&nbsp;&nbsp; &nbsp;On the contrary, we propose here to work with the C-scan representation of the ultrasonic signal, i.e. the spatial ultrasonic data distribution at a fixed time. This representation hence provides a data representation closer to an image, in which the defect is clearly observable. This “image-like” representation is then more adapted for machine learning tools developed by the computer vision community. Using our experimental database, which we plan to make publicly available, we have developed several defects detection methods, from unsupervised and classic machine learning methods to deep learning approaches. In particular, we have investigated the use of a semantic segmentation network that provides a classification of the data at the pixel level, hence a localization of the defect within the material. In this paper, we compare the classification results of these methods and discuss their drawbacks and advantages.

Benefit of Neural Network for the Optimization of Defect Detection on Composite Material Using Ultrasonic Non Destructive Testing

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.