A deep learning-based approach for RGB Micro-LED defect detection using microscopic vision

Interface debonding between high-strength grouting materials and the inner surfaces of steel tubes in grouting jacket connections (GJCs), which have been widely employed in offshore wind turbine support structures, negatively affects their mechanical behavior. In this study, an interface debonding defect detection and localization approach for scaled underwater GJC specimens using surface wave measurements with piezoelectric lead zirconate titanate (PZT) actuation and sensing technology was validated experimentally. Firstly, GJC specimens with artificially mimicked interface debonding defects of varying dimensions were designed and fabricated in the lab, and the specimens were immersed in water to replicate the actual underwater working environment of GJCs in offshore wind turbine structures. Secondly, to verify the feasibility of the proposed interface debonding detection approach using surface wave measurements, the influence of the height and circumferential dimension of the debonding defects on the output voltage signal of PZT sensors was systematically studied experimentally using a one pitch and one catch (OPOC) configuration. Thirdly, a one pitch and multiple catch (OPMC) configuration was further employed to localize and visualize the debonding defect regions. An abnormal value analysis was carried out on the amplitude of the output voltage signals from PZT sensors with identical wave traveling paths, and the corresponding abnormal surface wave propagation paths were identified. Finally, based on the influence of interface debonding on the surface wave measurements mentioned above, the mimicked interface debonding defect was detected successfully and the region of debonding was determined with the intersection of the identified abnormal wave travelling paths. The results showed that the mimicked debonding defect can be visualized. The feasibility of this method for interface debonding defect detection in underwater GJCs was confirmed experimentally. The proposed approach provides a novel non-destructive debonding defect detection approach for the GJCs in offshore wind turbine structures.

In the competitive world of the food industry where companies have to offer quality products, quality control is essential. However, it could become expensive, especially if it is a manual process. Its automation then becomes an excellent opportunity for a company. The objective of this research is to find out whether it is possible to carry out quality control of open mouth bag sealings on industrial bagging systems using deep learning. In this paper, we propose a three-step approach: data collection, data classification, and supervised classification learning. The first step is to collect images of sealings of open mouth bags. We created a line-scan based prototype and placed it on a production line to harvest a large amount of data. Image processing is then applied to clean the data. The next step is the classification of the data to identify classes of defects and labeling of these data. Finally, supervised classification learning makes it possible to implement quality control. We propose an architecture based on convolutional neural networks for image classification of open mouth bags. Our approach gives very encouraging results for the realization of quality control of an industrial bagging system.

Homoscedasticity of homogeneous flat surface products is a necessary condition for a high quality product. The quality of homogeneous flat surface products, like paper sheets, steel slabs, textiles, and glasses, plays a crucial role in raising the profile of the manufacturing companies. This paper presents a new approach for defect detection using the Levene's test, which is used for testing the homogeneity of variances of samples drawn from the same population. It is assumed that the variances of samples taken from the same population are equal. Occurrence of defects results in a Levene's test measure that is higher than some critical value indicating that the null hypothesis of equal variances is rejected. Noise immunity of the proposed technique is ensured through pre-filtering the fabric image using the Wiener filter that is an edge preserving filter. The robustness to variations of the sliding window size and the structures of fabric is analyzed. The major advantages of the proposed technique are the low computational complexity and noise immunity while maintaining high accuracy.

Internal defects detection and classification in hollow cylindrical surfaces using single shot detection and MobileNet

BGNN4VD: Constructing Bidirectional Graph Neural-Network for Vulnerability Detection

Natural gas vehicles have been widely used around the world as a mode of green transportation. The defect detection method for a vehicle cylinder, which is the energy storage component of a natural gas vehicle, is essential for ensuring safety. Current vehicle cylinder testing methods are relatively simple and cannot quickly and accurately detect internal defects in the winding layer of the cylinder. In this study, the effect of winding layer defects on the temperature behavior of cylinders during filling was studied using finite element simulation and infrared thermal imaging technology. The results show that a defect in the winding layer acts as a source of thermal resistance that hinders heat transfer. The more serious the defect is, the greater the hindrance of the heat transfer is. According to this principle, an evaluation model based on the R parameter was developed for the temperature changes due to defects, and a method for quantitatively correlating the maximum R value with the defect depth was established, thereby providing a practical approach for the rapid detection and evaluation of defects in cylinder winding layers.

Integral images-based approach for fabric defect detection

In the recent past, quadratic frequency-modulated thermal wave imaging (QFMTWI) has been advanced with a chirp z-transform (CZT)-based processing approach to facilitate enhanced subsurface anomaly detection, depth quantification and material property estimation with enhanced depth resolution. In the present study, the applicability of CZT-based phase analysis for foreign object defect detection in a structural steel sample using QFMTWI is validated through finite element-based numerical modelling rather than experimental verification due to limited available resources. Furthermore, the enhanced defect detection capability of the CZT phase approach is qualitatively compared with the frequency- and time-domain phase approaches using the defect signal-to-noise ratio (SNR) as a quality metric. Also, an empirical relationship between the observed phases and the thermal reflection coefficient is obtained, which recommends the CZT phase as a prominent approach for foreign material defect detection.

We present a new approach for non-destructive evaluation (NDE) in homogeneous and isotropic metallic objects which contain defects at unknown depths based on single scan and multi frequency excitation. As known, there is an optimum frequency for each depth of defect. Finding the depth of an unknown defect requires us to find the optimal frequency. In conventional single frequency methods, the optimal frequency is obtained by applying a wide range of frequencies to the system separately and comparing the corresponding results in a time-consuming process. Conventional multi frequency inspections were introduced to obtain more information about test specimens. There are two ways to apply multiple excitation frequencies to the test sample. First way is to apply them sequentially to a single excitation coil which is still time consuming and the second method is to apply them simultaneously to multiple coils. In simultaneous excitation the testing time is shorter, but its processing of phase and amplitude signals is still challenging. We propose and examine the capability of the multi frequency eddy current (MFEC) method to detect the depth of hidden cracks in a shorter time period; using our HTS SQUID gradiometer-based system. The measurements are performed in a noisy environment, and a planar double-D shaped printed-circuit-board coil is used as the excitation coil. Comparing with the results of the consecutive single frequency excitation measurements, the obtained depths of the flaws using MFEC method was confirmed.

A systematic literature review on software defect prediction using artificial intelligence: Datasets, Data Validation Methods, Approaches, and Tools

This paper proposes an optimal approach for real-time defect detection in industrial assembly lines using Convolutional Neural Networks (CNN’s) implemented on ARM Cortex – M microcontrollers. Although CNNs are very accurate when detecting visual defects, their computational complexity makes them difficult to implement on devices with limited resources. To address this issue, techniques such as model pruning, quantization and memory optimizations tailored to the ARM Cortex -M microcontroller will be used. This results in an average inference time of 80-100 ms per image and the system achieves 92% accuracy which makes it suitable for industrial applications.

Benchmarking a multi-layer approach and neural network architectures for defect detection in PBF-LB/M

Object detection and recognition is an essential task in computer vision with numerous real-world applications such as surveillance, self-driving cars, and robotics. In recent years, deep learning-based approaches have significantly improved the accuracy and speed of object detection and recognition. The You Only Look Once version 3 (YOLOv3) algorithm is a popular deep learning-based approach that can detect and recognize objects in real-time. The Common Objects in Context (COCO) dataset is a large-scale dataset with over 330,000 labeled images and more than 2.5 million object instances, making it a popular choice for object detection and recognition tasks. In this paper, we propose a deep learning-based approach for real-time object detection and recognition using the YOLOv3 architecture and COCO dataset. We evaluate our approach based on several performance metrics, including mean average precision (mAP), frames per second (FPS), total object detection time, object detection accuracy, false positive rate, number of detected objects, and mean intersection over union (mIoU). Our results show that our approach achieves a mean average precision of 0.76 on the COCO dataset and a real-time performance of 40 frames per second on a single GPU. Additionally, our approach achieves an object detection accuracy of 93.5%, a false positive rate of 6.5%, and a mean intersection over union of 0.65. Our proposed approach shows promising results for real-time object detection and recognition and can be applied to various real-world applications.

Remote sensing is a key tool to derive glacier surface velocities but existing mapping methods, such as cross-correlation techniques, can fail where surface properties change temporally or where large velocity variations occur spatially. High-resolution datasets, such as UAV imagery, offer a promising solution to tackle these issues and to study small-scale glacier dynamics, but new workflows are required to handle such data. Therefore, we tested the potential of new deep learning-based image-matching algorithms for deriving glacier surface velocities across the ablation area of a glacier with strong spatial variability in surface velocities (<5 m/yr to >100 m/yr) and substantial changes in surface properties between image acquisitions. For a thorough comparison of state-of-the-art methods and sensors, we applied three different techniques (cross-correlation using geoCosiCorr3D, feature tracking with ORB using SeaIceDrift and the new deep learning-based method using ICEpy4D) and three different platforms (Sentinel-2, PlanetScope, UAVs) to estimate glacier surface velocities. Results showed lowest errors for velocities derived with the deep learning-based approach applied to UAV imagery (RMSE = 2.17 m/yr, R2 = 0.99), followed by cross-correlation using Sentinel-2 images (RMSE = 21.0 m/yr, R2 = 0.59) and the deep learning-based approach with PlanetScope data (RMSE = 21.28 m/yr, R2 = 0.36). Cross-correlation with geoCosiCorr3D resulted in comparably high errors with the UAV dataset (RMSE = 36.22 m/yr, R2 = 0.24), whereas ORB-based feature tacking showed lowest performance with all sensors. Spatial patterns of computed velocities indicate that applying existing cross-correlation methods for areas with regular displacements or low glacier velocities yields suitable results on UAV data, but innovative deep learning-based approaches are required for resolving rapid changes in velocities or in surface properties. This novel method benefits from improved keypoint detection and matching through training using neural networks and data characterized by challenging geometries, outlier minimization and more robust descriptors by applying cross-attention layers. We conclude that continued development of deep learning-based feature tracking approaches for glacier velocity computations may substantially improve UAV-based velocity derivations applied to challenging situations. This method is able to deliver reliable displacement data in situations where traditional methods fail, which implies a new level of detail in understanding and interpreting glacier dynamics.

Automated fibre placement (AFP) machine in-process layup inspection is essential for composite structure fabrication efficiency and quantity improvement. The laser profilometer scanning technique is one of the automated in-process inspection techniques. Laser profilometers can capture a large number of intensive high-resolution point clouds during the layup. To date, few studies have been published about processing point clouds at high speed for in-process layup defect detection and layup feature measurement. In this study, an algorithm called the cross-sectional line-processing algorithm was proposed, and a testbed was constructed to validate the algorithm. The algorithm processes each laser line captured by the laser profilometer and clusters the processed results together. Finally, the defect features and layup features can be segmented and recognized from the collected point cloud. Layup defect types can be recognized, and the dimensions of the layup features can be measured. The experimental results show that this approach can process at most 200 laser lines (about 160 000 points) per second, and the overall defect type recognition accuracy rate reaches about 78%, which meets basic AFP in-process inspection requirements.