The Effect of Different Flaw Data to Machine Learning Powered Ultrasonic Inspection

Abstract

Previous research (Li et al., Understanding the disharmony between dropout and batch normalization by variance shift. CoRR abs/1801.05134 (2018). http://arxiv.org/abs/1801.05134arXiv:1801.05134) has shown the plausibility of using a modern deep convolutional neural network to detect flaws from phased-array ultrasonic data. This brings the repeatability and effectiveness of automated systems to complex ultrasonic signal evaluation, previously done exclusively by human inspectors. The major breakthrough was to use virtual flaws to generate ample flaw data for the teaching of the algorithm. This enabled the use of raw ultrasonic scan data for detection and to leverage some of the approaches used in machine learning for image recognition. Unlike traditional image recognition, training data for ultrasonic inspection is scarce. While virtual flaws allow us to broaden the data considerably, original flaws with proper flaw-size distribution are still required. This is of course the same for training human inspectors. The training of human inspectors is usually done with easily manufacturable flaws such as side-drilled holes and EDM notches. While the difference between these easily manufactured artificial flaws and real flaws is obvious, human inspectors still manage to train with them and perform well in real inspection scenarios. In the present work, we use a modern, deep convolutional neural network to detect flaws from phased-array ultrasonic data and compare the results achieved from different training data obtained from various artificial flaws. The model demonstrated good generalization capability toward flaw sizes larger than the original training data, and the effect of the minimum flaw size in the data set affects the $$a_{90/95}$$ a 90 / 95 value. This work also demonstrates how different artificial flaws, solidification cracks, EDM notch and simple simulated flaws generalize differently.

Automated defect detection in digital radiography of aerospace welds using deep learning

“…Even when using only one image for generating training patches, the combined data augmentation achieved a good a 90/95 result. With pure virtual flaw augmentation, used by Koskinen et al [27], there can be a loss of some subtle features related to the combination of location and signal, not perfectly captured by virtual flaw, which is then alleviated by mixing defects in their original locations and new locations. Moreover, the distribution of flaw sizes and shapes is slightly skewed from the original, the effect of which is reduced by having half of the data set follow the original distribution.…”

“…This is a simpler approach in comparison to simulation methods [13,33] that require each defect type to be modeled in a representative way. Koskinen et al [27] also found simulation to yield limited generalization. In the case of cracks, the combined method did not differ from pure virtual, since no cracks from the original material were present in the training data set, but they were rather used for validation.…”

“…Virkkunen et al [50] used the virtual flaw technology as data augmentation for a deep learning classifier for ultrasonic testing (UT) data. Koskinen et al [27] compared different types of virtual flaws and simulated flaws as training data for UT, finding that virtual flaws outperformed purely simulated data, and that the simulated data on its own was insufficient for generalization. The advantage of virtual flaw is that is generates variability to both background and flaw signals, while having both the defects and background drawn from real data.…”

“…Previous studies by Virkkunen et al [50] and Koskinen et al [27] using virtual flaw augmentation on UT data had a limited amount of real flaws in their data sets, so a pure virtual flaw approach was used. However, the material in this study provides a larger representation of defect locations in its original form, without possibility for artefacts or skewed distribution that may be present in the virtual flaw process.…”

Acoustic Resonance Testing of Small Data on Sintered Cogwheels

“…For more details, we refer to [ 18 , 19 , 20 ] and the references therein. For each gear wheel, the raw signal of acoustic response that goes through a preamplifier was recorded by two receivers (channels 1 and 2) with a sampling rate of 1041.67 kHz with respect to ten different positions:…”