Defect Image Recognition and Classification for Eddy Current Testing of Titanium Plate Based on Convolutional Neural Network

Abstract

In the actual production environment, the eddy current imaging inspection of titanium plate defects is prone to scan shift, scale distortion, and noise interference in varying degrees, which leads to the defect false detection and even missed inspection. In view of this problem, a novel image recognition and classification method based on convolutional neural network (CNN) for eddy current detection of titanium plate defects is proposed. By constructing a variety of experimental conditions and collecting defect signals, the characteristics of eddy current testing (ECT) signals for titanium plate defects are analyzed, and then the convolution structure and learning parameters are set. The structural characteristics of local connectivity and shared weights of CNN have better feature learning and characterization capabilities for titanium plate defect images under scan shift, scale distortion, and strong noise interference. The results prove that, compared with other deep learning and classical machine learning methods, the CNN has a higher recognition and classification accuracy for the defect eddy current image of the titanium plate in the complex detection environment.

Depth Evaluation for Metal Surface Defects by Eddy Current Testing Using Deep Residual Convolutional Neural Networks

“…Firstly, thirty volunteers, who had no experience operating an ECT device, were invited to scan the defects, hence introducing a great variety of uncertainties, as composed to some research, e.g. [20]- [23], where an automatically controlled movement was harnessed to scan defects. Secondly, lift-off signals were deliberately collected and labeled, so that the classifier should be able to differentiate lift-offs and defects.…”

“…A Deep Belief Network (DBN), constructed by stacking multiple Restricted Boltzmann Machines, was exploited in [22] so as to, from the EC scan images of the defects on the surface of a Titanium sheet, extract features that were then fed to a least-square SVM algorithm to classify the defects. The dataset was also evaluated in [23] with a plain Convolutional Neural Network (CNN), which, in contrast to [22] where the feature extractor and classifier were separate, was trained end-to-end from EC signals to the classification labels of defects. Classification accuracy increased to 99.79% from the 96% in [22].…”

“…Instead, properly designed loss function and network architecture played important roles. In [23]- [26], CNN was used which was one of the most popular networks in DL research. Nonetheless, the adopted CNNs were wide and shallow, which was at variance with the 'deep' feature of modern neural networks.…”

Depth Evaluation for Metal Surface Defects by Eddy Current Testing using Deep Residual Convolutional Neural Networks

“…A Deep Belief Network was exploited in [10] so as to, from the EC scan images of the defects on the surface of a Titanium sheet, extract features that were then fed to a least-square SVM algorithm to classify the defects. The dataset was also evaluated in [11] with a plain Convolutional Neural Network (CNN), which, in contrast to [10] where the feature extractor and classifier were separate, was trained end-to-end. In [12], an encoder-decoder CNN, named EddyNet, was proposed aiming at learning an inverse model, which predicted a crack profile given an EC signal.…”

“…Specifically, the numbers of training samples in [12]- [14] were more than twenty thousand, while those in [4]- [9] were mostly a few hundreds. In [11]- [14], CNN was used which was one of the most popular networks in DL research. Nonetheless, the adopted CNNs were wide and shallow, which was at variance with the 'deep' feature of modern neural networks.…”