Augmented Ultrasonic Data for Machine Learning

Abstract

Flaw detection in non-destructive testing, especially in complex signals like ultrasonic data, has thus far relied heavily on the expertise and judgement of trained human inspectors. While automated systems have been used for a long time, these have mostly been limited to using simple decision automation, such as signal amplitude threshold.The recent advances in various machine learning algorithms have solved many similarly difficult classification problems, that have previously been considered intractable. For non-destructive testing, encouraging results have already been reported in the open literature, but the use of machine learning is still very limited in NDT applications in the field. Key issue hindering their use, is the limited availability of representative flawed data-sets to be used for training.In the present paper, we develop modern, very deep convolutional network to detect flaws from phased-array ultrasonic data. We make extensive use of data augmentation to enhance the initially limited raw data and to aid learning. The data augmentation utilizes virtual flaws -a technique, that has successfully been used in training human inspectors and is soon to be used in nuclear inspection qualification. The results from the machine learning classifier are compared to human performance. We show, that using sophisticated data augmentation, modern deep learning networks can be trained to achieve superhuman performance by significant margin.

Generative adversarial network with object detector discriminator for enhanced defect detection on ultrasonic B-scans

“…However, a B-scan usually consists of hundreds of A-scans which further aggravates the problem of lack of data. Developed algorithms for defect detection can be divided into three groups related to data representation being used; A-scans [14,15,16,17,18,19,20,7,21,22,8,23], B-scans [24,25,6,5] and C-scans [26,27]. The A-scan analysis is the most researched group of all which is also related to the data problem.…”

“…B-scans keep the geometrical coherence of the defect which leads to a better noise immunity [24]. However, the analysis of B-scans can only be seen in a few works [5,6]. In [5] two popular deep learning object detection models, YOLOv3 and SSD, have been used for defect detection.…”

“…In [5] two popular deep learning object detection models, YOLOv3 and SSD, have been used for defect detection. In [6] a deep learning detector has been tested on augmented images, but with only three defects in the specimen block. Regarding C-scans, in [26] a method based on the comparison of the scan with a reconstructed reference image has been made.…”

“…Automated analysis has long been used in many NDT systems. However, so far it has been limited to classical decision-making algorithms such as amplitude threshold [6]. Complex data such as the one from ultrasonic inspection makes it hard to develop an automated analysis.…”

“…The automated analysis could make the process much faster and reliable. There have been some attempts in developing an automated UT analysis [7,8,9,6,5], but very few of them involve using deep learning and modern deep convolutional neural networks (CNNs) on B-scans. The prerequisite for using deep learning is a large, annotated dataset.…”