Journal of Nondestructive Evaluation最新文献

This study explores the sensitivity of the Magnetic Barkhausen Noise (MBN) technique in detecting various types and degrees of burning in a single sample, which is similar to what occurs in industrial processes. Using flat grinding with an aluminum oxide wheel on hardened AISI 1045 steel, eight samples with a ground area of 115 mm x 7 mm were created, varying only the ae parameter. In some samples, the effect of generating different levels of burning was observed, starting at one end (grinding wheel entrance) without damage and gradually increasing the damage until the opposite end (grinding wheel exit) with the presence of high levels of burning and the identification of a thick white layer. Results indicated that the MBN_RMS (root mean square value of the MBN signals) parameter can identify varying burning levels caused by overtempering and rehardening. Burning gradients were clearly detected by MBN and confirmed by metallographic analyses. When the white layer is generated continuously on the surface, the MBN_RMS parameter adequately tracks the variation in its thickness, varying in an inversely proportional manner.

Shearography, an effective non-destructive testing tool, is widely employed for detecting defects in composite materials. It detects internal defects by detecting deformation anomalies, offering advantages such as full-field, non-contact measurement, and high accuracy. Defect size is a critical parameter determining structure performance stability and service life. However, manual inspection is the primary method for defect size measurement in this technique, leading to inefficiency and low accuracy. To address this issue, this study established a defect recognition and high-precision automatic measurement method based on the U-Net deep learning network. First, a high-precision one-time calibration method for all system parameters was developed. Second, U-Net was employed to segment the measured image, identifying defect location and subimage. Finally, defect size was accurately calculated by combining calibration parameters and segmented defect subimage. The proposed method yielded a measurement error of less than 5% and a real-time dynamic detection rate of 14 fps, demonstrating potential for automated quantitative defect detection.

The possibility of assessing the thickness of the hardened layer on the surface of AISI 321 metastable austenitic steel, subjected to frictional treatment with a sliding indenter under various normal loads, using the magnetic Barkhausen noise method and the eddy-current method is investigated. The production of hardened layers of different thicknesses is simulated by stepwise electrolytic etching. The results of the non-destructive methods were compared to those obtained by the microhardness method to determine the thickness of the hardened layer. It is shown that the thickness of the hardened layer can be assessed using the eddy-current method and the magnetic Barkhausen noise method. However, the eddy-current method is preferable. This is because, in addition to sensitivity to the ferromagnetic phase, it is also sensitive to the level of defectiveness of the γ-phase. At the same time, it is necessary to take into account in the test method that the thickness of the hardened layer determined by the non-destructive methods is less than that determined by the microhardness method.

This research explores the process of generating artificial training data for the detection and classification of defective areas in X-ray computed tomography (XCT) scans in the agricultural domain using AI techniques. It aims to determine the minimum detectability limit for such defects through analyses regarding the Probability of Detection based on analytic XCT simulations. For this purpose, the presented methodology introduces randomized shape variations in surface models used as descriptors for specimens in XCT simulations for generating virtual XCT data. Specifically, the agricultural sector is targeted in this work in terms of analyzing common degradation or defective areas in rice products. This is of special interest due to the huge biological genotypic and phenotypic variations occurring in nature. The proposed method is demonstrated on the application of analyzing rice grains for common defects (chalky and pore areas).

In order to measure non-destructively the ³⁵Cl in cementitious materials a special NMR setup was developed. Besides ³⁵Cl also quasi-simultaneously both ²³Na and ¹H can be measured. This setup is built around a 4.7 T wide bore superconducting magnet. The present results show that using this setup we can measure non-destructively the ³⁵Cl, ²³Na and ¹H in ordinary Portland cement samples. Using the present setup ³⁵Cl, ²³Na and ¹H profiles can be measured over a longer period of time, hence giving for example the possibility to look at the dynamic binding process of ³⁵Cl and ²³Na during hydration, as is demonstrated. Moreover, the measurement time with the present setup gives the possibility to look at the dynamics processes like, for example, the NaCl solution absorption as is demonstrated, showing NMR can be used for non-destructive evaluation.

Functional performance of Expose Aggregate Concrete Pavement (EACP) such low tire-pavement noise and higher skid resistance are noticeable due to long-term durability, are influenced by wavelength and mean texture depth (MTD). EACP surface macrotexture is characterized by the MTD and exposed aggregate number (EAN) due to a higher correlation between wavelength and the EAN. Normally, the EAN is manually estimated which needs much human effort and is time-consuming. Recently, deep learning of computer vision has been employed for aiding human counting tasks in different condition. Mostly, many state-of-the-arts for counting are conducted by using RGB image which is color image. Regarding the counting techniques used for EAN, it is a challenging task to deal with some issues such as aggregate is some occluded and similar coloring to the background. Because the aggregate shows the peak characteristic, the depth value may benefit in improving the recognition. This additional information may be useful since it can be display distinguishable color between the object and background. Therefore, this study aims to evaluate the combination of RGB image and depth information, knowns as RGB-D image, for counting the EAN by adapted Faster RCNN deep learning model with four channel input images. The RGB-D dataset was newly constructed for training and testing implemented model. The result shows the accuracy slightly improve by 5% by using RGB-D compared to RGB. However, they both achieve similar MAE and RMSE. Therefore, it gives the valuable information for EAN counting. Both image datasets are acceptable for counting the EAN with a given condition.

Structural health monitoring of carbon fiber components is critical in high-stakes applications such as aerospace and prosthetics. Carbon fiber’s exceptional mechanical properties demand precise defect detection to ensure safety and longevity. This paper reviews recent advancements in monitoring carbon fiber aircraft structures using a custom optical coherence tomography (OCT) imaging system. This innovative system integrates hyperspectral imaging with automated classifiers to detect and classify both surface and subsurface defects, including roughness and cracks. By employing OCT with magnitude and quantitative phase imaging algorithms, the study introduces methods for detailed three-dimensional visualization of material defects. The high-resolution capabilities of the OCT system enable accurate and automated crack detection, enhancing reliability in critical applications. The paper also addresses challenges in deploying these advanced systems in practical scenarios, such as integration with existing maintenance protocols and data interpretation. It explores the potential of combining OCT with other non-destructive evaluation techniques to improve monitoring accuracy. These advancements contribute to more reliable, non-invasive monitoring of carbon fiber structures, with significant implications for safety and performance in various industries.

The surface of complex textured ceramic tiles contains numerous defects that exhibit low contrast with the background, making them easily confused with the textured background during detection. Traditional defect detection algorithms and convolutional neural networks are prone to texture interference in the defect detection of complex textured ceramic tiles, resulting in high false detection rates and missed detection rates. Inspired by the human eye’s ability to find surface defects on smooth objects against a high-light background, this paper proposes a new method for detecting surface defects of complex textured tiles. This method uses the high-light area generated by the reflection of the light source as the background for detecting textured tile defects, thereby increasing the threshold difference between the defect and the background and highlighting the defect. This method translates the position of the textured tiles horizontally and captures images while the reflection of the strip light source covering the surface of the tiles is in motion, thereby acquiring several tile images with light source reflections. Subsequently, after intercepting the images of the highlight areas covered by the light source reflection, the RANSAC algorithm is used to match the characteristic corners of these images, and after rigid splicing, a complete image of the textured tiles with the highlight area as the background is obtained. Finally, defects on textured tiles can be extracted through threshold segmentation and morphological filtering. Experimental results indicate that this method can ignore complex texture interference on ceramic tiles and achieve rapid detection of defects in textured ceramic tiles.

In the last few years, infrared long pulse thermography (LPT) has attained high reliability and accuracy in the non-destructive inspection of low-thermally conductive materials such as carbon fibre reinforced polymer (CFRP) composites. However, to date, research investigations of LPT have been conducted on standardised and controlled material flaws such as flat bottom holes. Non-standardised defects in CFRPs are more common in real-life operations and, because of different nature, dimensions and complex shapes, their detection poses a significant challenge. This paper provides an in-depth analysis of LPT combined to advanced statistical and machine learning-based image processing tools for detection of non-standardised damage in CFRP composites. Statistical methods such as skewness and kurtosis, and machine learning algorithms such as principal component analysis and Fuzzy-c clustering were used to post-process thermal LPT signals. Damage scenarios that are likely to occur during manufacturing and in-service operations were analysed in terms of defect mapping characteristics using the signal-to-noise ratio and the Tanimoto criterion. Experimental results revealed that Fuzzy-c and LPT produced superior damage inspection performance.

Differentiating between wheat species poses a significant challenge to the Indian grain industry. Visual inspection of wheat species has drawbacks, including inconsistency, low throughput, and labor intensiveness. In this study, near-infrared hyperspectral imaging (NIR-HSI) was utilized in conjunction with a deep learning approach to achieve precise predictions of wheat at the species level. A dataset comprising 40 different varieties from four Indian wheat species, namely Triticum aestivum (T. aestivum), Triticum durum (T. durum), Triticum dicocccum (T. dicoccum), and Triticale, was prepared using a NIR-HSI system that encompassed the wavelength ranging from 900–1700 nm. The imbalanced dataset is a common problem in the classification task, making it harder for the classifier to classify minority class data correctly. To address this issue, oversampling techniques such as Synthetic Minority Oversampling Technique (SMOTE) and Adaptive Synthetic sampling (ADASYN) were employed. For the classification task, a 1D Convolutional Neural Network (1D-CNN), a 1D-ResNet, and four traditional machine learning models: Naive Bayes (NB), K-Nearest Neighbor (KNN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) are utilized and compared. The performance of these models was assessed using both imbalanced and balanced datasets. The 1D-CNN model outperformed traditional machine learning models, achieving impressive test accuracies of 98.25% and 98.43% with the SMOTE and ADASYN approach, respectively. These findings underscore the efficacy of NIR-HSI in conjunction with an end-to-end 1D-CNN and oversampling techniques as a reliable and efficient method for the rapid, accurate, and nondestructive identification of various wheat species. The code is available at https://github.com/nitintyagi007-iitr/Wheat_species_classification