Tiny Defects Research Articles

Classification of Defective and Non-Defective Products Using Convolutional Neural Networks in Quality Control

The arrival of convolutional neural networks (CNNs) has enhanced the progress of computer visualisation from many fields. However, most of the CNNs are rely on GPUs (graphics processing units) that could needthe large computations and it requires more cost to develop the setup. Therefore, most of the manufacturers haven’t used the CNNs to inspect the defective items in theirfield. The researcher has developed a compact CNN-based model that not only achieves high performance on tiny defect inspection but can be run on low- frequency CPUs (central processing units) in this paper. This experiments indicate CNNs can be compact and hardware-friendly for future applications in the automated surface inspect (ASI)in the selected manufacturing field.

Anomaly Candidate Extraction and Detection for automatic quality inspection of metal casting products using high-resolution images

This paper proposes an AI-based detection method called ACED (Anomaly Candidate Extraction and Detection) that can find tiny defects efficiently by inspecting the surface of cast manufactured products through high-resolution images. A major challenge is to minimize false detection to a level competent to skilled human inspectors. Also, high-resolution imaging is necessary to capture meaningful information about tiny flaws, but it may cause overfitting of AI models and increase computational burdens. We have designed ACED in a few steps to overcome the challenges: first, we create image partitions using factory-defined inspection area information and create regions of interest containing the most abnormal spots found by a convolutional autoencoder. Next, we filter out regions with minor flaws using the specification of significant flaws, where the information is available from a production line. Finally, we train an ensemble of convolutional neural networks to classify the defect candidates as normal or defective. We implemented and tested the proposed method ACED for two cast-manufactured automobile parts in a test production line of our collaborator’s factory. Our experimental results showed that ACED was particularly better in reducing false positive detection, outperforming other AI-based detection methods by ×10.5∼×43.0 in terms of accurate rates and ×5.3∼×18.2 in F1-scores, depending on the tested products. Also, ACED was competent to human inspectors, achieving ×1.9∼×4.0 performance gains in F1-scores. We also provide analyses of execution time, resource utilization, and inspection time reduction of factory workers, showing the potential of ACED to be used in actual production lines.

GBH-YOLOv5: Ghost Convolution with BottleneckCSP and Tiny Target Prediction Head Incorporating YOLOv5 for PV Panel Defect Detection

Photovoltaic (PV) panel surface-defect detection technology is crucial for the PV industry to perform smart maintenance. Using computer vision technology to detect PV panel surface defects can ensure better accuracy while reducing the workload of traditional worker field inspections. However, multiple tiny defects on the PV panel surface and the high similarity between different defects make it challenging to accurately identify and detect such defects. This paper proposes an approach named Ghost convolution with BottleneckCSP and a tiny target prediction head incorporating YOLOv5 (GBH-YOLOv5) for PV panel defect detection. To ensure better accuracy on multiscale targets, the BottleneckCSP module is introduced to add a prediction head for tiny target detection to alleviate tiny defect misses, using Ghost convolution to improve the model inference speed and reduce the number of parameters. First, the original image is compressed and cropped to enlarge the defect size physically. Then, the processed images are input into GBH-YOLOv5, and the depth features are extracted through network processing based on Ghost convolution, the application of the BottleneckCSP module, and the prediction head of tiny targets. Finally, the extracted features are classified by a Feature Pyramid Network (FPN) and a Path Aggregation Network (PAN) structure. Meanwhile, we compare our method with state-of-the-art methods to verify the effectiveness of the proposed method. The proposed PV panel surface-defect detection network improves the mAP performance by at least 27.8%.

An Efficient Channel Attention-Enhanced Lightweight Neural Network Model for Metal Surface Defect Detection

There are problems of low model detection accuracy, low detection speed and difficulty in deploying online inspection in industrial surface defect detection relying on deep learning object detection algorithms. In order to effectively solve this problem, an efficient channel attention-enhanced lightweight neural network model named as EMV2-YOLOX is proposed in this paper. The algorithm incorporates the ECA module into the lightweight backbone extraction network MobileNetV2 to achieve adaptive adjustment of channel information weights, which can improve the extraction capability of the algorithm. The YOLOX model is also introduced to enhance the model’s identification and localization of tiny defects. The improved algorithm can guarantee the model’s accuracy and improve the model detection performance, as well as the carrying capacity of hardware devices. The experimental results show that the highest accuracy is achieved on the GCT10 and NEU public defect datasets with mean Average Precision values of 0.86 and 0.68, respectively, which is higher than the accuracy of the EMV2yoloV4 model. The parametric model number is only 10.24[Formula: see text]M in size, and the detection rate is 54.25[Formula: see text]f/s, which is the highest performance in embedded devices. EMV2-YOLOX, combined with the attention mechanism, can efficiently extract the location and semantic information of hard-to-detect defects and plays a vital role in the intelligent detection methods.

Optical System for Thin Transparent Surface Inspection

Automated inspection system now a day is a must in mass production lines. To optimize the abilities of the inspection system using computer vision (CV) technology in terms of accuracy and speed the input data should be as good as possible. The quality of input data for a computer vision system plays a very important role in this story. We can define a high-quality image as one that is characterized by clarity, sharp contrast, exceptional resolution, and a high system resolution (um/pixel). The ability to highlight defects in contrast to non-defective areas within a sample is a critical aspect of good input data, as it facilitates precise defect detection. When input data meets these criteria, computer vision systems can operate with enhanced accuracy and speed, ultimately improving the overall efficiency of quality control processes. All above requirements can be archived just in case considering to the optical system (OPS) of CS system. The design of an optical system will be based on samples as per material, dimension, defect descriptions and requested system speed. There's many types of inspected samples and defects that will lead to the OPS's design being impossible or cannot be optimized to fit with requested requirement. One of them is the thin and transparent surface with tiny defects as: crack, dent, chip off... with the size of couple of 100 µm. This paper introduces an OPS specifically tailored for inspecting storage disk surfaces before the coating stage in production lines, addressing these challenges. The goal is to enhance precision and efficiency in defect detection, contributing to advancements in automated inspection systems for specialized industrial applications.

A New Multiscale Multiattention Convolutional Neural Network for Fine-Grained Surface Defect Detection

Surface defect detection (SDD) is essential in industry to ensure the quality of products. Fine-grained SDD has been developed to cope with the high intra-class and low inter-class differences in defect samples. However, the sizes and appearances of the defects are various, making it still challenging to identify the defects accurately. To overcome this drawback, this paper proposes a new multi-scale multi-attention convolutional neural network (MSMA-SDD) for fine-grained SDD. Firstly, a new multi-scale CNN network is developed, which uses features from different layers to match the defects with different sizes. Secondly, a new multi-attention module is proposed, which generates compact attention map to assist MSMA-SDD focusing on the tiny defects. Thirdly, the multi-scale and multi-attention based MSMA-SDD is adopted in the fine-grained SDD task. The proposed MSMA-SDD is conducted on three datasets. The experimental results show that MSMA-SDD is superior to the current most advanced method, and the accuracy are 100%, 99.59% and 99.57% respectively, which has verified its potential in the SDD field.

A context-aware progressive attention aggregation network for fabric defect detection

Fabric defect detection plays a critical role for measuring quality control in the textile manufacturing industry. Deep learning-based saliency models can quickly spot the most interesting regions that attract human attention from the complex background, which have been successfully applied in fabric defect detection. However, most of the previous methods mainly adopted multi-level feature aggregation yet ignored the complementary relationship among different features, and thus resulted in poor representation capability for the tiny and slender defects. To remedy these issues, we propose a novel saliency-based fabric defect detection network, which can exploit the complementary information between different layers to enhance the representation features ability and discrimination of defects. Specifically, a multi-scale feature aggregation unit (MFAU) is proposed to effectively characterize the multi-scale contextual features. Besides, a feature fusion refinement module (FFR) composed of an attention fusion unit (AFU) and an auxiliary refinement unit (ARU) is designed to exploit complementary important information and further refine the input features for enhancing the discriminative ability of defect features. Finally, a multi-level deep supervision (MDS) is adopted to guide the model to generate more accurate saliency maps. Under different evaluation metrics, our proposed method outperforms most state-of-the-art methods on our developed fabric datasets.

Convolution With Rotation Invariance for Online Detection of Tiny Defects on Magnetic Tile Surface

Convolution With Rotation Invariance for Online Detection of Tiny Defects on Magnetic Tile Surface

A Tiny Defect Detection System for Tire Mold Surfaces Based on Consecutive Frames

The purpose of the study is to achieve the precise location of each defect on the surfaces of the customized tire mold. Traditional image processing systems typically face significant challenges when processing images, including noise, distortions, and tiny objects, resulting in low computing efficiency and detection accuracy. To address these concerns, a comprehensive defect detection system is developed. The system consists of an inspection platform and a surface defect detection method. In order to achieve highly sensitive detection, an inspection platform and a novel filtering mechanism are proposed. The inspection platform is suitable for various customized tire molds and can be adjusted according to different situations. The effect of noise is effectively reduced by the novel filtering mechanism using a continuous image. Defect features are also prevented from being filtered out. In addition, to remove unnecessary parts of the image, U <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -Net is adopted to avoid detecting defective features outside the mold. Finally, a tire mold dataset—STM is set up for training and evaluating U <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> -Net. The proposed method achieves 83.6% precision and 87.9% recall, which is the best performance among all comparisons. Furthermore, this system can provide defect details for subsequent analysis and automatic defect repair.

MCTSR: A super‐resolution method for defects in gas‐insulated switchgear X‐ray images based on multi‐scale contextual transformers

AbstractAs the core equipment of transmission and distribution hubs, the operational status of gas‐insulated switchgear (GIS) is closely linked to the safety of the power system. Recently, X‐ray digital imaging technology has been extensively used in GIS equipment fault detection. However, the X‐ray image of GIS is blurred, which is not conducive to the detection of tiny defects. Thus, a super‐resolution method for GIS X‐ray images based on multi‐scale context transformers is proposed in this study, namely MCTSR. Firstly, a second‐order image degradation model is introduced to generate GIS X‐ray low‐resolution images that more closely resemble the real world. Secondly, a contextual transformer gate module is constructed to improve attention to tiny defects in GIS X‐ray images. Thirdly, a U‐Net discriminator network based on multi‐scale contextual transformers is intended to enrich the information of the generated images. Finally, the proposed discriminator is combined with the existing generator to compose a super‐resolution method applicable to GIS X‐ray images. The experimental results demonstrate that the method outperforms other methods in peak signal‐to‐noise ratio and structural similarity on the constructed GIS X‐ray image dataset. In addition, the output image of the proposed method facilitates the subsequent defect detection.

Lightweight and Strong Ceramic Network with Exceptional Damage Tolerance.

Lightweight materials such as porous ceramics have attracted increasing attention for applications in energy conservation, aerospace and automobile industries. However, porous ceramics are usually weak and brittle; in particular, tiny defects could cause catastrophic failure, which affects their reliability and limits the potential use greatly. Here we report a SiC/SiO2 nanowire network constructed from numerous well-bonded SiC nanowires coated by a biphasic structure consisting of amorphous SiO2 and nanocrystal SiC. The as-obtained SiC/SiO2 nanowire network is lightweight (360 ± 10 mg cm-3), mechanically strong (compressive strength of 16 MPa), and damage-tolerant. The high strength of the network is attributed to the biphasic mixed structure of the binding coating which can restrict the deformation of nanowires upon compression. The lightweight and strong SiC/SiO2 nanowire network shows potential for engineering applications in harsh environments.

Experimental and Numerical Analyses of a Novel Magnetostatic Force Sensor for Defect Inspection in Ferromagnetic Materials

An innovative magnetostatic force sensor consisting of a laser source, a tiny cantilever beam, and a small permanent magnet was developed and used for defect inspection in ferromagnetic samples in the present article. The penetrating zone within a ferromagnetic material under the magnetic field provided by a permanent magnet was called the magnetic sensing zone (MSZ), and surface or internal defects within the MSZ were inspected by measuring the change in the magnetostatic force. This magnetostatic force could be calculated by the Maxwell tensor integrating over the surface and interface of a ferromagnetic material. Numerical and experimental results demonstrated that this sensor was reliable and could precisely inspect the defects of different sizes in ferromagnetic samples. In summary, the sensor proposed in this paper has the potential for industrial applications to detect surface and sub-surface tiny defects on ferromagnetic steel thin sheets, such as the zinc slag defect of hot galvanized sheets, cracks on cold-rolled sheets, and the ferromagnetic oscillation marks of continuous casting.

MicroCrack-Net: A Deep Neural Network With Outline Profile-Guided Feature Augmentation and Attention-Based Multiscale Fusion for MicroCrack Detection of Tantalum Capacitors

The microcrack defect in the terminal electrode of the tantalum capacitor seriously affects the service life of the capacitor. However, the microcrack displays low contrast and large noises in the image. Meanwhile, the width of the microcrack is small, and the area in the image is tiny, which brings additional challenges to the detection of microcracks on the terminal electrodes. To improve the performance of detecting microcracks, this article proposes a detection model with outline profile-guided feature augmentation and attention-based multiscale fusion, titled with MicroCrack-Net. In this method, the microcrack outline is utilized to guide the lossless feature extraction to strengthen unobvious microcrack features without pooling operation to avoid information loss of tiny defects during the layer-by-layer processing in the convolutional neural network. A gradual attention mechanism is proposed to attract the attention of the detection model to local defect areas in the image with noises. Multiscale feature extraction and fusion are designed to process the data in tensors of the same size. The experimental results demonstrated that the proposed model is effective and has superior performance over VGG16, resNet50, resNext50, DenseNet, Racki-Net, and SegDecNet in detecting microcracks.

Digital Twin and Data-Driven Quality Prediction of Complex Die-Casting Manufacturing

Digital twin and data-driven technologies provide an idea for realizing complex product quality prediction. Aiming at the issues of real-time visual monitoring, operating status analysis, and quality prediction in complex die-casting intelligent manufacturing, a digital twin and data-driven quality prediction architecture is proposed. The virtual-real interaction digital twin of die-casting manufacturing cells is established. The collaborative working mode of physical cells, virtual cells, and real-time monitoring is constructed to predict product quality. The learning method of die-casting parameter data and appearance defect data is proposed to realize the real-time quality prediction in die-casting process and the appearance defect quality prediction after processing, respectively. The data preprocessing and XGBoost-based learning method is proposed for real-time quality prediction of die-casting process. A single-shot refinement neural network for aluminum casting tiny defects detection (Refine-ACTDD) based on deep learning is proposed to solve the high-precision defect detection problems of small appearance defects of complex castings and large interference of complex background. Taking the complex aluminum die-casting as an example, the applications of quality prediction are verified. The method provides a new technical approach for high-precision quality prediction of complex die-casting manufacturing.

A review of non-destructive methods for the detection tiny defects within organic insulating materials

In recent years, solid organic materials such as silicone rubber and epoxy resin have been widely used in electrical equipment due to their excellent insulation properties. However, as a result of manufacturing and design flaws as well as aging issues during operation, the insulating materials in the linked state no longer fit tightly and tiny structural defects (defect size less than 10 mm) develop, such as debonding at the composite interface, pores or cracks within the insulating material, etc. Tiny defects are prone to partial discharges and breakdowns, compromising the safety of high-voltage power equipment, particularly when subjected to strong electric fields. Therefore, it is necessary to carry out non-destructive testing (NDT) for such tiny defects. Such defects are small in size, easily buried in the material, and even some are wrapped in metal, which in turn requires very high detection accuracy, but traditional methods are difficult to achieve, so NDT technologies for tiny defects within insulating materials have become a research hotspot in the field of electric power in recent years. This paper firstly introduces the sources of tiny defects in solid organic insulating materials for electrical equipment. Secondly, the harm caused by structural defects is elaborated. Finally, emerging NDT methods and their advantages and limitations in defect detection are described in detail. The review aims to provide the reader with a comprehensive overview of most of the NDT techniques used in the detection of tiny defects within solid organic insulating materials for electrical equipment and their most salient features.

Unsupervised defect detection with patch-aware mutual reasoning network in image data

Current defect detection studies in the industrial fields mainly adopt supervised strategies, which require large amounts of annotated defective samples to achieve superior results. However, it is hard to meet such large-scale data requirements in the actual industrial scenarios where defects commonly exhibit intensive scarcity and huge intraclass variation. To address the above limitations, we propose a novel Patch-aware Mutual Reasoning Network (PMRN) that utilizes only the prior knowledge of non-defective samples for defect detection. Concretely, a patch-aware mutual reasoning module and a spatial shuffle perception module are devised to reason mutual dependencies and explore dislocations relationships. Besides, an adaptive soft gated anomaly measurement function is developed to calculate reconstruction deviations, which can soft control the information flow according to the complexity of the current scenario. Extensive experimental results (five benchmark datasets: average AUC of 0.951) demonstrate that the PMRN can accurately detect defects with complex backgrounds and weak textures using only non-defective data for training, thus relieving the dependence on large-scale annotated defective samples. Future work will focus on the unsupervised detection of tiny defects.

A lightweight and efficient model for surface tiny defect detection

A lightweight and efficient model for surface tiny defect detection

Preparation and enhanced oxidation behavior of microalloyed Mo5SiB2 alloy at 1300 °C

Microalloyed Mo5SiB2 alloys were fabricated by spark plasma sintering (SPS). The effects of different sintering temperatures and yttrium additions on the microstructure and high-temperature oxidation behavior of the alloys were systematically investigated. For the optimum Mo-13Si-25B-0.2Y alloy sintered at 1750 °C, adding small amounts of rare-earth yttrium refined the grains to an average size of ~400 nm without reducing the Mo5SiB2 content. During oxidation at 1300 °C, the “fishlike” Y2Si2O7 and Y2O3 phases were formed and anchored on the surface of the oxide layers, promoting the oxidation of silicon and accelerating the formation and coverage of the oxide protective film. Moreover, yttrium promoted the in situ deformation of a self-generated protective borosilicate glass film under high-temperature oxidation by accelerating Si oxidation. With prolonged oxidation time, viscous flow occurred to repair the internal cracks and tiny defects caused by the volatilization of MoO3 and prevented the inward diffusion of oxygen, as evidenced by a low oxidation rate constant of 0.95 mg2 cm−4 h−1 at 1300 °C for 50 h in the Mo-13Si-25B-0.2Y alloy.

Modified Mini-Keystone Flaps for Coverage of Tiny Volar Pulp Defects of the Fingertips in Cases with Missing Amputation Skin Stumps: A Retrospective Study.

This study aimed to demonstrate the expanding versatility of keystone flap reconstruction in fingertips. Fifteen patients who underwent the modified mini-keystone flap reconstruction for tiny volar pulp defects of the fingertip between September 2020 and February 2021 were included in this study (average age: 43.4 ± 13.52 years, range: 19–61 years). Patient data were retrospectively collected from their medical records. The two-point discrimination test was used to evaluate the degree of sensory recovery. All defects were successfully covered with the modified mini-keystone flap. The defect sizes ranged from 0.5 cm × 1 cm to 1.2 cm × 2.0 cm, and the flap sizes ranged from 0.7 cm × 1.5 cm to 1.5 cm × 3.0 cm. Although one patient showed a small distal margin maceration, all flaps survived fully. The overall outcomes were favorable at the mean follow-up period of 5.73 ± 0.79 months. We suggest that the modified mini-keystone flap technique is a promising alternative modality for covering tiny volar pulp defects of the fingertip, with few complications and favorable outcomes.

Anterior cranial fossa osseous arteriovenous fistula of the crista galli with bone erosion: patient series

BACKGROUNDAlthough an anterior cranial fossa dural arteriovenous fistula (ACFdAVF) is thought to have a fistula on the dura near the olfactory groove, the detailed angioarchitecture remains unreported.OBSERVATIONSIn case 1, a 65-year-old man was found to have an asymptomatic ACFAVF. His computed tomography angiography (CTA)-maximum intensity projection (MIP) showed the shunt point in the crista galli (CG), with the intradural drainer penetrating the destroyed bone of the CG. In case 2, a 78-year-old man had a past history of intracerebral hemorrhage and was found to have an ACFAVF. The rotational angiography (RA)-MIP showed the intraosseous fistula in the CG with the drainer passing through a tiny bone defect of the CG. In case 3, a 35-year-old man was investigated for epilepsy. The RA-MIP showed an osseous arteriovenous fistula (AVF) in the anterior cranial base, with the drainer penetrating the skull osteolytic site. In case 4, a 73-year-old woman was found to have an asymptomatic ACFAVF. Her RA-MIP showed the osseous AVF with the drainer penetrating the CG with bone erosion.LESSSONSAll patients were diagnosed with anterior cranial fossa osseous AVF rather than dAVF, with bone erosion in the CG. These findings should be noted at the time of diagnosis and treatment.