Defects In Region Research Articles

A Novel Defect Quantification Method Utilizing Multi-Sensor Magnetic Flux Leakage Signal Fusion

In the assessment of pipeline integrity using magnetic flux leakage (MFL) detection, it is crucial to quantify defects accurately and efficiently using MFL signals. However, in complex detection environments, traditional defect inversion methods exhibit low quantification accuracy and efficiency due to the complexity of their algorithms or excessive reliance on a priori knowledge and expert experience. To address these issues, this study presents a novel defect quantification method based on multi-sensor signal fusion (MSSF). The method employs a multi-sensor probe to fuse the MFL signals under multiple lift-off values, enhancing the diversity of defect information. This enables defect-opening profile recognition using the characteristic approximation approach (CAA). Subsequently, the MSSF method is based on a 3D magnetic dipole model and integrates the structural features of multi-sensor probes to develop an algorithm. This algorithm iteratively determines the defect depth at multiple data acquisition points within the defect region to obtain the maximum defect depth. The feasibility of the MSSF quantification method is validated through finite element simulation and physical experiments. The results demonstrate that the proposed method achieves accurate defect quantification while enhancing efficiency, with the number of iterations for each defect depth calculation point consistently requiring fewer than 15 iterations. For rectangular metal loss, perforation, and conical defects, quantification errors are less than 10%, meeting practical inspection requirements.

Intelligent identification of defective regions of voids in tunnels based on GPR data

Quantitatively detecting voids behind tunnel linings presents significant challenges in identifying the range of width and depth. This paper develops an innovative method for identifying defective regions of voids based on Ground Penetrating Radar (GPR) data. This method involves three steps: Firstly, the void-identifying-feature-set (VIFS) is constructed by extracting the Amplitude peak (AT), Signal energy (ET), and Amplitude peak of the first intrinsic mode function (IMF1) component (AH) of every A-scan signal. Secondly, the Support Vector Machine (SVM) is utilized to identify defect signals and normal signals, contributing to the width identification of void in the horizontal direction. Thirdly, an innovative Three-Stage-Boundary-Extraction (TSBE) algorithm is proposed to identify the depth range of voids in the vertical direction. Experimental results conducted on both field data and simulated data demonstrated that the Intersection over Union (IOU) value and consumption time of three groups of GPR data (Data I, Data II, and Data V) are 0.739 and 0.888 s, respectively. The average IOU and consumption time of the TSBE algorithm are 0.739 and 0.058 s, respectively.

Experimental Evaluation of the Possibility of Through Defects Detection in Thermally Expanded Graphite Workpieces by Acoustic Method

The paper presents the results of evaluating the possibility of detecting defects (e.g., through holes) in workpieces made of thermally expanded graphite used as sealing materials using a non-contact air-coupled acoustic inspection system. The deviation of the acoustic signal amplitude when it passes through a sample of thermally expanded graphite in defective and defect-free regions is used as an informative parameter. The obtained dependences of the change in the amplitude of the passed signal through the sample on the size of the artificial through hole, density and thickness of the object showed a significant influence of these parameters on the detectability of the defect, which should be taken into account in the development of inspection technologies and in the process of its implementation. The change in the amplitude of the signal passed through the object directly depends on the increase in the parameters of the sample (density, thickness) and the size of the through hole, which increases the sensitivity of acoustic control and detectability of defects.On the basis of the obtained dependences the criteria for rejection of workpieces can be developed at the stage of production or immediately before manufacturing of their products. The influence of defect positioning relative to the acoustic axis of the transmitter-receiver system is shown, which leads to a decrease in the informative parameter, and consequently to a decrease in sensitivity. This dependence is characteristic for all thicknesses and densities of objects, as well as for all sizes of artificial through holes when control is carried out immediately after defect formation. When acoustic inspection is carried out some time after the defect formation, the maximum of this dependence may shift due to the shift of the amount of substance to fill the defect cavity, which is confirmed by microscopy. The obtained results can be used in the development of workpieces acoustic control parameters.

Segmentation-assisted classification model with convolutional neural network for weld defect detection

Detecting weld defects in battery trays is crucial for the safety of new energy vehicles. Existing methods for weld surface defect detection, relying on traditional computer vision algorithms and convolutional neural networks with substantial image-level labeled data, face challenges in accurately identifying small defects, especially with limited samples. To address these issues, we developed an innovative Segmentation-Assisted Classification with Convolutional Neural Networks (SACNN) model. SACNN integrates a common feature extraction subnet, a segmentation subnet enhanced by a multi-scale feature fusion module, and a classification subnet specifically designed for precise defect detection. A joint loss function co-trains the segmentation and classification subnets using both image-level and pixel-level labels, enhancing the model’s ability to accurately detect small defect regions. Our model demonstrates notable improvement, achieving accuracy gains ranging from 2% to 18% compared to existing state-of-the-art methods, with an overall accuracy of 94.09% on an industrial dataset of battery tray welds. To further evaluate the generalization capability of our model, we evaluated it on the publicly available Magnetic Tile dataset, achieving state-of-the-art results in this challenging context. Additionally, we conducted comprehensive ablation studies to validate the contribution of each component in our approach and utilized visualization techniques to enhance the interpretability of our model. These advancements represent a significant contribution to the state of the art in aluminum alloy weld defect detection.

Deep learning of 3D point clouds for detecting geometric defects in gears

Geometric integrity directly impacts the functionality, reliability, and safety of final manufactured products, making the qualification of parts based on measurements of their geometry a fundamental quality control activity in modern manufacturing. Recent advancements in three-dimensional (3D) metrology technologies have enabled fine-scale inspection of geometric integrity characterized by dimensional accuracy, surface quality, and shape conformity. However, the widespread adoption of high-resolution 3D metrology in manufacturing faces some significant challenges posed by the inherent data structures of 3D point clouds such as high dimensionality, unstructured nature, and sparsity in defective regions. To address these challenges, this paper first creates a “MFGNet-gear” dataset, which is a scalable and comprehensive benchmark dataset comprising 12 part designs with four quality classes for each design. Subsequently, we develop a deep learning model adapted from the PointNet++ architecture to enable automated, end-to-end analysis of 3D point clouds. The model can be configured for different decision-making tasks including part design classification and multi-class geometric defect detection. Implementations of the proposed model on the “MFGNet-gear” dataset achieve accuracies up to 100% in classifying gear designs and up to 85% in four-class quality inspection. Additionally, we systematically investigate the impacts of measurement resolution and precision on the classification performance through a series of case studies. The obtained results highlight the potential of using deep learning methods for automated analysis of 3D point clouds for a variety of quality control tasks beyond gear manufacturing. This study also proposes future research directions, including the development of new deep learning architectures specifically designed for manufacturing 3D point clouds and strategies for adaptive measurement planning.

Multi-scale feature reconstruction network for industrial anomaly detection

Unsupervised anomaly detection techniques, which operate without prior knowledge of anomalies, have garnered significant attention in industrial inspection due to their adaptability and generalization. Therefore, knowledge-based computer vision techniques have been broadly applied to identify unusual image patterns. However, real-time industrial applications present challenges such as limited anomalous samples, inadequate defect knowledge, and complex background textures. These factors lead to difficulties in accurately identifying defect regions, and conventional auto-encoder networks often struggle to overcome these issues.To address these limitations, we propose a multi-scale feature reconstruction (MSFR) network specifically designed for domain shift scenarios. Our approach employs a pyramidal vision transformer network (PVTN) to reconstruct multi-scale feature maps, capturing discriminative features at various scales. Additionally, a pre-trained module extracts multi-level features at the same scale, and a dedicated feature matching module enhances accuracy by improving the alignment probability between features. The MSFR strategy surpasses conventional auto-encoders by filtering pixel-level information at multiple depths.Empirical evaluations were conducted using benchmark datasets such as MVTec AD and AeBAD-S. Furthermore, an extensive ablation study demonstrates the effectiveness and viability of the proposed MSFR approach for industrial anomaly detection tasks. The experimental results show that the proposed model significantly outperforms recent approaches, making it highly suitable for real-world industrial applications, particularly in manufacturing.

3D-Mapping and Manipulation of Photocurrent in an Optoelectronic Diamond Device.

Establishing connections between material impurities and charge transport properties in emerging electronic and quantum materials, such as wide-bandgap semiconductors, demands new diagnostic methods tailored to these unique systems. Many such materials host optically-active defect centers which offer a powerful in situ characterization system, but one that typically relies on the weak spin-electric field coupling to measure electronic phenomena. In this work, charge-state sensitive optical microscopy is combined with photoelectric detection of an array of nitrogen-vacancy (NV) centers to directly image the flow of charge carriers inside a diamond optoelectronic device, in 3D and with temporal resolution. Optical control is used to change the charge state of background impurities inside the diamond on-demand, resulting in drastically different current flow such as filamentary channels nucleating from specific, defective regions of the device. Conducting channels that control carrier flow, key steps toward optically reconfigurable, wide-bandgap optoelectronics are then engineered using light. This work might be extended to probe other wide-bandgap semiconductors (SiC, GaN) relevant to present and emerging electronic and quantumtechnologies.

A customised ConvNeXt-SCC network: integrating improved principal component analysis with ConvNeXt to enhance tire crown defect detection

ABSTRACT Defects such as bubbles and rubber shortages occasionally occur during vehicle tire production. Tiny defects in the crown gradually expand under load, ultimately causing tire failure. However, machine-vision-based crown defect detection faces challenges like limited datasets, varying groove patterns, low colour contrast, and small defect sizes. Therefore, this article integrates principal component analysis (PCA) and deep learning techniques for the visual detection of tire crown defects. A dataset of tire crown defects suitable for deep learning was constructed using an improved-PCA method and composite sampling. We propose the ConvNeXt-Shallow Conv CBAM (ConvNeXt-SCC) model, which includes an improved convolutional block attention module (CBAM) in a shallow stacked structure. And we propose an efficient and accurate method called Improved Selective Search (ISS) to detect and locate tire crown defective regions. The ISS pre-selected box generation method, enhanced with a priori knowledge filtering module, was used alongside a classification model to achieve online inference detection of large-size tire crown images. The experimental results show that the performance indices of the proposed ConvNeXt-Shallow Conv CBAM (ConvNeXt-SCC) model significantly improve compared to the original ConvNeXt-T. The defect detection and localisation method based on ISS meets the online inspection requirements of tire production.

Halofilins as emerging bactofilin families of archaeal cell shape plasticity orchestrators

Bactofilins are rigid, nonpolar bacterial cytoskeletal filaments that link cellular processes to specific curvatures of the cytoplasmic membrane. Although homologs of bactofilins have been identified in archaea and eukaryotes, functional studies have remained confined to bacterial systems. Here, we characterize representatives of two families of archaeal bactofilins from the pleomorphic archaeon Haloferax volcanii, halofilin A (HalA) and halofilin B (HalB). HalA and HalB polymerize in vitro, assembling into straight bundles. HalA polymers are highly dynamic and accumulate at positive membrane curvatures in vivo, whereas HalB forms more static foci that localize in areas of local negative curvatures on the outer cell surface. Gene deletions and live-cell imaging show that halofilins are critical in maintaining morphological integrity during shape transition from disk (sessile) to rod (motile). Morphological defects in ΔhalA result in accumulation of highly positive curvatures in rods but not in disks. Conversely, disk-shaped cells are exclusively affected by halB deletion, resulting in flatter cells. Furthermore, while ΔhalA and ΔhalB cells imprecisely determine the future division plane, defects arise predominantly during the disk-to-rod shape remodeling. The deletion of halA in the haloarchaeon Halobacterium salinarum, whose cells are consistently rod-shaped, impacted morphogenesis but not cell division. Increased levels of halofilins enforced drastic deformations in cells devoid of the S-layer, suggesting that HalB polymers are more stable at defective S-layer lattice regions. Our results suggest that halofilins might play a significant mechanical scaffolding role in addition to possibly directing envelope synthesis.

Electric current-induced directional slip of dislocation and grain boundary ordering

The evolution of lattice dislocations in metals can be induced by electric current. However, the effect of electric current on the interaction between lattice dislocations and grain boundaries (GBs) remains unclear. Herein, we investigate the evolution of lattice dislocations and grain boundary dislocations (GBDs) induced by electric current in M50 steel via in-situ TEM characterization, density functional theory (DFT) calculations, and molecular dynamics (MD) simulations. Our findings reveal that electric current induces the directional slip of lattice dislocations towards GBs, while the reduction of GB contrast is attributed to the decomposition and reorganization of GBDs. DFT calculations demonstrate that electric current enhances the force on atoms in defect regions. These in-situ TEM observations are well reproduced by MD simulations incorporating electric current, which include a virtual temperature rise at defect regions to reflect the electric current force on defect atoms. Based on these observations, we propose a novel GB accommodation model under electric current. These results provide valuable insights into the mechanisms by which electric current tailors lattice dislocation behavior and enhances GBDs ordering in metallic materials.

Detection and diagnosis of diabetic eye diseases using two phase transfer learning approach.

Early diagnosis and treatment of diabetic eye disease (DED) improve prognosis and lessen the possibility of permanent vision loss. Screening of retinal fundus images is a significant process widely employed for diagnosing patients with DED or other eye problems. However, considerable time and effort are required to detect these images manually. Deep learning approaches in machine learning have attained superior performance for the binary classification of healthy and pathological retinal fundus images. In contrast, multi-class retinal eye disease classification is still a difficult task. Therefore, a two-phase transfer learning approach is developed in this research for automated classification and segmentation of multi-class DED pathologies. In the first step, a Modified ResNet-50 model pre-trained on the ImageNet dataset was transferred and learned to classify normal diabetic macular edema (DME), diabetic retinopathy, glaucoma, and cataracts. In the second step, the defective region of multiple eye diseases is segmented using the transfer learning-based DenseUNet model. From the publicly accessible dataset, the suggested model is assessed using several retinal fundus images. Our proposed model for multi-class classification achieves a maximum specificity of 99.73%, a sensitivity of 99.54%, and an accuracy of 99.67%.

Accurate Inspection and Super-Resolution Reconstruction for Additive Manufactured Defects Based on Stokes Vector Method and Deep Learning

Defects in additive manufacturing processes are closely related to the mechanical and physical properties of the components. However, the extreme conditions of high temperatures, intense light, and powder during the manufacturing process present significant challenges for defect detection. Additionally, the high reflectivity of metallic components can cause pixels in image sensors to become overexposed, resulting in the loss of many defect signals. Thus, this paper mainly focuses on proposing an accurate inspection and super-resolution reconstruction method for additive manufactured defects based on Stokes vector and deep learning, where the Stokes vectors, polarization degree, and polarization angles of the inspected defects are effectively utilized to suppress the high reflectivity of metallic surfaces, enhance the contrast of defect regions, and highlight the boundaries of defects. Furthermore, a modified SRGAN model designated SRGAN-H is presented by employing an additional convolutional layer and activation functions, including Harswish and Tanh, to accelerate the convergence of the SRGAN-H network and improve the reconstruction of the additive manufactured defect region. The experiment results demonstrated that the SRGAN-H model outperformed SRGAN and traditional SR reconstruction algorithms in terms of the images of Stokes vectors, polarization degree, and polarization angles. For the scratch and hole test sets, the PSNR values were 33.405 and 31.159, respectively, and the SSIM values were 0.890 and 0.896, respectively. These results reflect the effectiveness of the SRGAN-H model in super-resolution reconstruction of scratch and hole images. For the scratch and hole images chosen in this study, the PSNR values of SRGAN-H for single image super-resolution reconstruction ranged from 31.86786 to 43.82374, higher than the results obtained by the pre-improvement SRGAN algorithm.

A preliminary study of the mechanical properties of 3D-printed personalized mesh titanium alloy prostheses and repair of hemi-mandibular defect in dogs.

This study is a preliminary investigation exploring the mechanical properties of three-dimensional (3D)-printed personalized mesh titanium alloy prostheses and the feasibility of repairing hemi-mandibular defects. The ANSYS 14.0 software and selective laser melting (SLM) were used to produce personalized mesh titanium alloy scaffolds. Scaffolds printed using different parameters underwent fatigue property tests and scanning electron microscopy (SEM) of the fracture points. Models of hemi-mandibular defects (encompassing the temporomandibular joint) were created using beagle dogs. Freeze-dried allogeneic mandibles or 3D-printed personalized mesh titanium alloy prostheses were used for repair. Gross observation, computed tomography (CT), SEM, and histological examinations were used to compare the two repair methods. The prostheses with filament diameters of 0.5 and 0.7 mm could withstand 14,000 times and >600,000 cycles of alternating stresses, respectively. The truss-structure scaffold with a large aperture and large aperture ratio could withstand roughly 250,000 cycles of alternating forces. The allogeneic mandible graft required intraoperative shaping, while the 3D-printed mesh titanium alloy prostheses were personalized and did not require intraoperative shaping. The articular disc on the non-operated sides experienced degenerative changes. No liver and kidney toxicity was observed in the two groups of animals. The 3D-printed mesh titanium alloy prostheses could effectively restore the shape of the mandibular defect region and reconstruct the temporomandibular joint.

Enhancing bone repair through improved angiogenesis and osteogenesis using mesoporous silica nanoparticle-loaded Konjac glucomannan-based interpenetrating network scaffolds

We have fabricated and characterized novel bioactive nanocomposite interpenetrating polymer network (IPN) scaffolds to treat bone defects by loading mesoporous silica nanoparticles (MSNs) into blends of Konjac glucomannan, polyvinyl alcohol, and polycaprolactone. By loading MSNs, we developed a porous nanocomposite scaffold with mechanical strengths comparable to cancellous bone. In vitro cell culture studies proved the cytocompatibility of the nanocomposite scaffolds. RT-PCR studies confirmed that these scaffolds significantly upregulated major osteogenic markers. The in vivo chick chorioallantoic membrane (CAM) assay confirmed the proangiogenic activity of the nanocomposite IPN scaffolds. In vivo studies were performed using Wistar rats to evaluate the scaffolds' compatibility, osteogenic activity, and proangiogenic properties. Liver and renal function tests confirmed that these scaffolds were nontoxic. X-ray and μ-CT results show that the bone defects treated with the nanocomposite scaffolds healed at a much faster rate compared to the untreated control and those treated with IPN scaffolds. H&E and Masson's trichrome staining showed angiogenesis near the newly formed bone and the presence of early-stage connective tissues, fibroblasts, and osteoblasts in the defect region at 8 weeks after surgery. Hence, these advantageous physicochemical and biological properties confirm that the nanocomposite IPN scaffolds are ideal for treating bone defects.

Insight into the initiation and propagation mechanism of desiccation cracking in clayey soil from DEM simulations

Desiccation cracking, a common natural phenomenon, significantly affects the mechanical and hydraulic properties of clayey soils. This study employs 3D Discrete Element Method (DEM) to investigate the initiation and propagation mechanisms of soil desiccation cracking. By integrating water evaporation at the soil surface and water transportation between soil grains into a traditional DEM model, we unveil a sequential and hierarchical pattern in the development of desiccation cracks. Initially, micro cracks proliferate and coalesce in regions of soil defects characterized by the lowest coordination numbers, leading to the genesis of primary cracks. Subsequently, a complex network of cracks propagates, governed by the interplay between cracking and the redistribution of the tensile stress field. As desiccation advances, the detachment of soil from the bottom wall and an increase in soil strength contribute to the stabilization of the crack network. Notably, the hierarchical nature of desiccation cracking is predominantly shaped by initial soil defects and subsequently refined by stress concentration within the soil clods, correlating with changes in the length-to-height ratio of the soil clods.

A bi-functional self-healing epoxy composite coating based on coordinated functionalized attapulgite/graphene oxide

Applications for graphene oxide (GO) in the field of corrosion protection will be severely constrained by the material’s poor dispersion and compatibility with epoxy resins, as well as the absence of self-healing corrosion protection in GO-based epoxy coatings. In this work, we developed a bi-functional 2-mercaptobenzothiazole (MBT) inhibitor-encapsulating GO-based nanocomposite (GAM) with pH-responsive release by implanting attapulgite (ATP) nanorods on the surface of GO. Additionally, GO-based nanocomposite can form covalent connections with epoxy resins and be evenly distributed in epoxy film via a ring opening process. Moreover, improved epoxy film compatibility and dispersion result in elevated mechanical strength and better impermeability, which prevent corrosion in both the active and passive modes. The corrosion resistance of GAM self-healing coating increased as the submersion period exceeded 24 h up to 0.54 M Ω cm2, resulting in the release of MBT inhibitor to form a passivation film at the defective region. Moreover, the improved charge transfer resistance (Rct) and the coating’s resistance to corrosive species (Rc) attribute to the self-healing ability of the GAM coating. The incorporation of GAM (2 %) nanocomposites into epoxy films enables long-term protection of AA2024-T3 by combining passive and self-healing corrosion prevention.

Relative kinetic stability of defect patterns in two-dimensional nematic liquid crystals with rectangular confinement.

Guiding and dynamically modulating topological defects are critical challenges in defect engineering of liquid crystals. Here, we employ molecular dynamics simulations to investigate the transition dynamics and relative kinetic stability of defect patterns in two-dimensional nematic Gay-Berne liquid crystals confined within rectangular geometries. We observe the formation of various defect patterns including long-axis, diagonal, X-shaped, composite, and bend configurations under different confinement conditions. The competition between boundary effects and the uniformity of nematic orientation induces the continuous realignment of liquid crystal molecules, facilitating the spatially continuous transformation of defect patterns over time. This transition involves changes in both defect types and their locations, typically initiating from defect regions. Furthermore, we demonstrate that the relative stability of these defect patterns can be effectively controlled by adjusting confinement parameters and external field conditions. Our findings provide fundamental insights into the transition kinetics of defect patterns in confined nematic liquid crystals, thereby enhancing our ability to manipulate topological defects for advanced applications.

In Vivo Effects of Biosilica and Spongin-Like Collagen Scaffolds on the Healing Process in Osteoporotic Rats.

Due to bioactive properties, introducing spongin-like collagen (SPG) into the biosilica (BS) extracted from marine sponges would present an enhanced biological material for improving osteoporotic fracture healing by increasing bone formation rate. Our aim was to characterize the morphology of the BS/SPG scaffolds by scanning electron microscopy (SEM), the chemical bonds of the material by Fourier transform infrared spectroscopy (FTIR), and evaluating the orthotopic in vivo response of BS/SPG scaffolds in tibial defects of osteoporotic fractures in rats (histology, histomorphometry, and immunohistochemistry) in two experimental periods (15 and 30days). SEM showed that scaffolds were porous, showing the spicules of BS and fibrous aspect of SPG. FTIR showed characteristic peaks of BS and SPG. For the in vivo studies, after 30days, BS and BS/SPG showed a higher amount of newly formed bone compared to the first experimental period, observed both in the periphery and in the central region of the bone defect. For histomorphometry, BS/SPG presented higher %BV/TV compared to the other experimental groups. After 15days, BS presented higher volumes of collagen type I. After 30days, all groups demonstrated higher volumes of collagen type III compared to volumes at 15days. After 30days, BS/SPG presented higher immunostaining of osteoprotegerin compared to the other experimental groups at the same experimental period. The results showed that BS and BS/SPG scaffolds were able to improve bone healing. Future research should focus on the effects of BS/SPG on longer periods in vivo studies.

Evaluating the mechanism and the capacitive properties of nickel boride electrodes for supercapacitor applications

This study investigates the electrochemical properties of nickel boride electrodes produced via a novel molten salt boron diffusion method; known as cathodic reduction and thermal diffusion-based boriding (CRTD-Bor). Pure nickel substrates with two different geometries (i.e. flat and foam) were borided in the borax-based electrolyte. The presence of nickel borides, composed of NiB, Ni4B3 and Ni2B with the order locations from the surface towards the matrix, was confirmed by X-ray diffractometry (XRD) and scanning electron microscopy (SEM) attached with electron probe microanalyzer (EPMA). The electrochemical behaviors of the produced electrodes were inspected through cyclic voltammetry (CV), galvanostatic charge-discharge (GCD), as well as impedance spectroscopy (EIS). The areal capacitances of the electrodes were calculated as 478 and 1385 mF/cm2 for flat and foam samples, respectively after the 2000th cycle of CV at 10 mV/s scan rate. The reaction kinetic followed the mixed-controlled model according to the Trasatti approach. The capacitive contribution (Co) measured in flat samples was 71 %; whereas, it was found as 40.6 % in the foam-shaped electrodes. Additionally, the mechanism behind the faradaic reactions was explored through X-ray photoelectron spectroscopy (XPS) and Fourier Transform Infrared Spectroscopy (FTIR) which were conducted both before and after the CV experiments. The obtained results showed that the possible mechanism was related to the formation of NiO bonds due to the oxygen saturation in the defective regions of the surfaces during electrochemical testing. Moreover, the importance of the electrode geometry on the performance was determined by the fact that the porous structure leads to more energy storage and favors a fast charge-discharge trend.

Visual defect obfuscation based self-supervised anomaly detection

Due to scarcity of anomaly situations in the early manufacturing stage, an unsupervised anomaly detection (UAD) approach is widely adopted which only uses normal samples for training. This approach is based on the assumption that the trained UAD model will accurately reconstruct normal patterns but struggles with unseen anomalies. To enhance the UAD performance, reconstruction-by-inpainting based methods have recently been investigated, especially on the masking strategy of suspected defective regions. However, there are still issues to overcome: (1) time-consuming inference due to multiple masking, (2) output inconsistency by random masking, and (3) inaccurate reconstruction of normal patterns for large masked areas. Motivated by this, this study proposes a novel reconstruction-by-inpainting method, dubbed Excision And Recovery (EAR), that features single deterministic masking based on the ImageNet pre-trained DINO-ViT and visual obfuscation for hint-providing. Experimental results on the MVTec AD dataset show that deterministic masking by pre-trained attention effectively cuts out suspected defective regions and resolves the aforementioned issues 1 and 2. Also, hint-providing by mosaicing proves to enhance the performance than emptying those regions by binary masking, thereby overcomes issue 3. The proposed approach achieves a high performance without any change of the model structure. Promising results are shown through laboratory tests with public industrial datasets. To suggest EAR be possibly adopted in various industries as a practically deployable solution, future steps include evaluating its applicability in relevant manufacturing environments.