Journal of Nondestructive Evaluation最新文献

The lack of real defect data samples has become a challenging problem for the effective application of deep learning networks in ultrasound target detection. This paper proposes a data augmented generative adversarial network (DCSGAN) aimed at overcoming the scarcity of welding ultrasonic defect data in training target detection networks. This network utilizes bilinear interpolation to expand the real data sample space, facilitating the extraction of high-dimensional defect spatial features through deeper networks. By obtaining a mixed dataset of generative data and real data, training and testing experiments are conducted on the object detection network. The experimental results demonstrate that the data augmentation method proposed in this paper effectively enhances the detection rate of ultrasonic welding defects in the target detection network, which has reference significance for similar application scenarios of ultrasonic defect detection.

The purpose of this paper is to conduct a thorough investigation of a stochastic eddy-current testing problem, when the geometric parameters of the system under study are characterized by uncertainty. Focusing on the case of subsurface defect detection, we devise reliable surrogates for the quantities of interest (QoI) based on the principles of the generalized polynomial chaos (PC) and using the orthogonal matching pursuit (OMP) solver to promote sparsity in the approximate models. In addition, a variance-based approach is implemented for the sequential construction of the necessary sample set, enabling more accurate estimation of the statistical metrics without imposing additional computational overhead. Apart from quantifying the inherent uncertainty, a sensitivity analysis is performed that assesses the impact of each geometric variable on the QoI, via the computation of Sobol indices. The efficiency of the OMP-PC algorithm is demonstrated in two variants of the subsurface-discontinuity problem, yielding at the same time useful conclusions regarding the properties of the stochastic outputs.

Titanium plate has a vital position in many industrial fields due to its outstanding characteristics, and the eddy current detection technology can quickly and non-destructively detect the defects of titanium plate, which is one of the crucial methods of titanium plate defect non-destructive testing. However, in the actual detection process, eddy current detection imaging is inevitably affected by noise interference to varying degrees, concerning the accuracy of defect classification recognition. Therefore, this study has proposed a titanium plate eddy current detection image classification method based on parallel sparse filtering and deep forest, which realizes the detection image's sparse feature extraction and defect classification. Firstly, the parallel sparse filtering network is constructed by adding another direction's feature extraction operation to the traditional sparse filtering. The parallel sparse filtering network extracts more comprehensive sparse features from the detection image. Secondly, a deep forest network is built, and the Bayesian optimization algorithm is used to optimize the network's hyperparameters. Finally, the deep forest network with optimized hyperparameters is used to classify and recognize the titanium plate defect eddy current detection images. The experimental results show that the proposed method has better feature representation and feature relevance learning ability, has higher classification accuracy under different levels of noise interference, with a classification accuracy increase of 3.09–40.65% compared to other conventional methods, and has better robustness and anti-noise ability.

Welding defects have a significant influence on welding quality and structural strength, and the rapid and accurate detection of welding defects is required. In order to achieve this goal, it is imperative to create corresponding high-quality datasets. However, capturing image information through a single sensor presents certain limitations. In this study, a magneto-optical imaging device and an infrared thermal imaging device were combined to collect images of resistance spot welding samples. The imaging principles of magneto-optical imaging device and the infrared thermal imaging device are discussed, and the possible factors affecting the imaging modes are analyzed. By synthesizing the 3D gray image, the gray histogram, and inherent image features, the imaging rules of magneto-optical image and the infrared image of resistance spot welding samples have been summarized. Under the guidance of these two image types and imaging modes, image enhancement technology has been utilized to optimize the quality of sample images. The Peak Signal to Noise Ratio (PSNR), Structural Similarity Index (SSIM), and Universal Image Quality Index (UIQI) indicators were used to evaluate the optimization quality of the enhanced images. Compared with Histogram Equalization (HE), the Gamma transform, Brightness Preserving Bi-Histogram Equalization (BPBHE), and the Digital Detail Enhancement (DDE) method, the scores of the enhanced infrared images showed improvement across all indicators. The magneto-optical image yielded the best results in the PSNR index, while the other two indices showed only moderate performance. The image dataset, enhanced with appropriate image enhancement techniques, can be utilized for further research into magneto-optical and infrared image information fusion and welding defect identification.

Time reverse modeling (TRM) is successfully applied to acoustic signals from a circular microphone array, for mapping of sudden cracking sound events. Numerical feasibility using synthetic acoustic sources followed by an experimental study with steel pendulum impacts on a steel plate is carried out. The mapping results from the numerical and experimental data are compared and verified using a delay-and-sum beamforming technique. Based on the feasibility and experimental study, a mapping error is estimated. In the main experimental study, cracking sound events obtained during a tensile test on a textile-reinforced concrete specimen are mapped with the TRM. The enhanced capability of the TRM to map simultaneously occurring cracking sound events along crack paths is demonstrated.

Inspection of defects in pipelines can be materialized by measuring ultrasonic guided waves the properties of which are conventionally analyzed with three-dimensional finite-element methods (FEM). They require complicated geometric discretization and memory consumption in a single analysis, thus are clumsy and limited to be used for field fast analysis. This work developed a systematic analytical approach to perform rapid assessment of mode-to-mode reflection for guided waves in a pipe owing to notches and used low-cost microprocessors for calculation. The mechanism of wave reflection was interpreted with the reciprocity theorem and a novel dynamic rigid-ring approximation. The theory successfully estimated the coefficient dependence of notch depths with an accuracy comparable to that obtained from a FEM, with the maximum error being less than 0.044. The developed algorithm was further implemented on an embedded system for computational complexity estimation. It shows the complete analytical theory sufficiently reduces computational memory and time cost by orders of magnitude while retaining good accuracy in determining mode-to-mode guided reflection by notches, which is a useful tool for practical pipeline applications.

The inspection of thick-section sandwich structures with skins around core materials such as honeycomb, balsa, and foam relies on low-frequency vibration techniques to identify defects through changes in amplitude or phase response. However, current industrial methods are often limited to detecting specific types of defects, potentially overlooking others. Moreover, these methods do not gather detailed information about the defect type or depth, as they only analyse a small portion of the available data instead of the full relevant response spectrum. This paper explores the scientific basis of using low-frequency vibration in the pitch-catch variant for defect detection in homogeneous solids, through analysis of the full relevant frequency spectrum (5–50 kHz). Defects in structures lead to reduced local stiffness and mass in the affected area, causing resonance in the layer above, resulting in amplified vibrations known as local defect resonance (LDR). In this work, an aluminium plate with a 40 mm diameter circular flat-bottomed hole (FBH) at a depth of 1 mm (representing a skin defect) is excited with a chirp signal of 5–50 kHz, and the response is monitored 17 mm away from the excitation point. Finite-element analysis (FEA) is used for the numerical model, addressing challenges in creating an accurate model. The process to optimise the numerical model and the reduce model-experiment error is outlined, including challenges such as the lack of knowledge of material damping. The study emphasizes the importance of modelling the probe’s stiffness and damping effects for achieving agreement between the model and experiment. After incorporating these effects, the maximum LDR frequency error decreased from approximately 3 kHz to less than 1 kHz. In addition, this study presents a method with the potential for defect classification through comparison to modelled responses. The minimum difference error was used to quantify the resonance frequencies’ error between the model and the experiment. Since the resonant frequencies are a function of the defect’s shape, size, and depth, a relatively low root mean squared (RMS) error across the resonance frequency error spectrum indicates the defect’s characteristics. Finally, defect detection and sizing using the pitch-catch probe are explored with a wide-band excitation signal and a line scan through the mid-plane of the defect. A method for defect sizing using a pitch-catch probe is presented and experimentally validated. Accurate defect sizing is achieved with the pitch-catch probe when the defect width is at least (ge ) twice the 17 mm pin-spacing of the probe.

Repeated inspections have been reported to improve the reliability of nondestructive inspection and can be evaluated by multiplying the likelihood function. However, repeated inspections conducted by a single inspector may not be independent, because the subsequent inspections may be influenced by previous inspection results. The probability of detection (POD) quantifies the sensitivity and reliability of an inspection system. In this study, eddy-current inspection data were used to assess the effect of repeated inspections on POD improvement. Specifically, repeated measures correlation (RMC) analysis was performed, which did not violate the assumption of independence to analyze intra-individual association, considering the nonindependence of repeated measures. Nonindependent repeated inspections performed using a combination of two datasets reduced the uncertainty in POD. Moreover, RMC yielded further improvements in POD and reduced the uncertainty.

X-ray cone-beam computed tomography (CBCT) is a powerful tool for nondestructive testing and evaluation, yet the CT image quality can be compromised by artifact due to X-ray scattering within dense materials such as metals. This problem leads to the need for hardware- and software-based scatter artifact correction to enhance the image quality. Recently, deep learning techniques have merged as a promising approach to obtain scatter-free images efficiently. However, these deep learning techniques rely heavily on training data, often gathered through simulation. Simulated CT images, unfortunately, do not accurately reproduce the real properties of objects, and physically accurate X-ray simulation still requires significant computation time, hindering the collection of a large number of CT images. To address these problems, we propose a deep learning framework for scatter artifact correction using projections obtained solely by real CT scanning. To this end, we utilize a beam-hole array (BHA) to block the X-rays deviating from the primary beam path, thereby capturing scatter-free X-ray intensity at certain detector pixels. As the BHA shadows a large portion of detector pixels, we incorporate several regularization losses to enhance the training process. Furthermore, we introduce radiographic data augmentation to mitigate the need for long scanning time, which is a concern as CT devices equipped with BHA require two series of CT scans. Experimental validation showed that the proposed framework outperforms a baseline method that learns simulated projections where the entire image is visible and does not contain scattering artifacts.

Machine learning in non-destructive testing (NDT) offers significant potential for efficient daily data analysis and uncovering previously unknown relationships in persistent problems. However, its successful application heavily depends on the availability of a diverse and well-labeled training dataset, which is often lacking, raising questions about the transferability of trained algorithms to new datasets. To examine this issue closely, the authors applied classifiers trained with laboratory Ground Penetrating Radar (GPR) data to categorize on-site moisture damage in layered building floors. The investigations were conducted at five different locations in Germany. For reference, cores were taken at each measurement point and labeled as (i) dry, (ii) with insulation damage, or (iii) with screed damage. Compared to the accuracies of 84 % to 90 % within the laboratory training data (504 B-Scans), the classifiers achieved a lower overall accuracy of 53 % for on-site data (72 B-Scans). This discrepancy is mainly attributable to a significantly higher dynamic of all signal features extracted from on-site measurements compared to laboratory training data. Nevertheless, this study highlights the promising sensitivity of GPR for identifying individual damage cases. In particular the results showing insulation damage, which cannot be detected by any other non-destructive method, revealed characteristic patterns. The accurate interpretation of such results still depends on trained personnel, whereby fully automated approaches would require a larger and diverse on-site data set. Until then, the findings of this work contribute to a more reliable analysis of moisture damage in building floors using GPR and offer practical insights into applying machine learning to non-destructive testing for civil engineering (NDT-CE).