Search

Anomaly detection is crucial in large-scale industrial manufacturing as it helps to detect and localize defective parts. Pre-training feature extractors on large-scale datasets is a popular approach for this task. Stringent data security, privacy regulations, high costs, and long acquisition time hinder the development of large-scale datasets for training and benchmarking. Despite recent work focusing primarily on the development of new anomaly detection methods based on such extractors, not much attention has been paid to the importance of the data used for pre-training. This study compares representative models pre-trained with fractal images against those pre-trained with ImageNet, without subsequent task-specific fine-tuning. We evaluated the performance of eleven state-of-the-art methods on MVTecAD, MVTec LOCO AD, and VisA, well-known benchmark datasets inspired by real-world industrial inspection scenarios. Further, we propose a novel method to create a dataset by combining the dynamically generated fractal images creating a "Multi-Formula" dataset. Even though pre-training with ImageNet leads to better results, fractals can achieve close performance to ImageNet under proper parametrization. This opens up the possibility for a new research direction where feature extractors could be trained on synthetically generated abstract datasets mitigating the ever-increasing demand for data in machine learning while circumventing privacy and security concerns. [ABSTRACT FROM AUTHOR]

We propose an unsupervised method for eigen tree hierarchies and quantisation group association for segmentation of corrosion in marine vessel hull inspection via camera images. Our unsupervised approach produces image segments that are examined to decide on defect recognition. The method generates a binary decision tree, which, by means of bottom-up pruning, is revised, and dominant leaf nodes predict the areas of interest. Our method is compared with other techniques, and the results indicate that it achieves better performance for true- vs. false-positive area against ideal (ground truth) coverage. [ABSTRACT FROM AUTHOR]

Sheet metal stamping is widely used for high-volume production. Despite the wide adoption, it can lead to defects in the manufactured components, making their quality unacceptable. Because of the variety of defects that can occur on the final product, human inspectors are frequently employed to detect them. However, they can be unreliable and costly, particularly at speeds that match the stamping rate. In this paper, we propose an automatic inspection framework for the stamping process that is based on computer vision and deep learning techniques. The low cost, remote sensing capability and simple implementation mean that it can be easily deployed in an industrial setting. A particular focus of this research is to account for the harsh lighting conditions and the highly reflective nature of products found in manufacturing environments that affect optical sensing techniques by making it difficult to capture the details of a scene. High dynamic range images can capture details of an environment in harsh lighting conditions, and in the context of this work, can capture highly reflective metals found in sheet metal stamping manufacturing. Building on this imaging technique, we propose a framework including a deep learning model to detect defects in sheet metal stamping parts. To test the framework, sheet metal 'Nakajima' samples were pressed with an industrial stamping press. Then optimally exposed, sequence of exposures, tone-mapped and high dynamic range images of the samples were used to train convolutional neural network-based detectors. Analysis of the resulting models showed that high dynamic range image-based models achieved substantially higher accuracy and minimal false-positive predictions. [ABSTRACT FROM AUTHOR]

Nondestructive and contactless online approaches for detecting defects in polymer films are of significant interest in manufacturing. This paper develops vision-based quality metrics for detecting the defects of width consistency, film edge straightness, and specks in a polymeric film production process. The three metrics are calculated from an online low-cost grayscale camera positioned over the moving film before the final collection roller and can be implemented in real-time to monitor the film manufacturing for process and quality control. The objective metrics are calibrated to correlate with an expert ranking of test samples, and results show that they can be used to detect defects and measure the quality of polymer films with satisfactory accuracy. [ABSTRACT FROM AUTHOR]

This paper proposes a novel system for measuring the deformation of anodes automatically in an electrolytic process, eliminating the need for manual intervention. The system employs cameras to acquire lateral perspective images of the anodes. These images are processed using a computer vision algorithm to give measurements of anode deformation, while considering potential errors arising from scene and object geometry. The system's results align with measurements conducted by operators across 71 anodes and were validated over 3900 more anodes in four different locations under different lightning and environmental conditions. This system improves efficiency, by automating a task that was previously carried out manually, and also safety by eliminating the operators need of handling heavy loads and operating in hazardous environments. • The system measures each anode pre- and post-maintenance, avoiding manual checks. • The system prevents handling heavy loads and operating in risky environments. • Tested and validated in real conditions, the system proves its performance. [ABSTRACT FROM AUTHOR]

Recent research topics have been placed on reinforcing security and safety measures within high-risk industries to protect both equipment and the environment. Numerous industries carry substantial implications for their surroundings. During significant incidents like chemical spills, or nuclear accidents, swiftly gathering precise and dynamic data poses a considerable challenge. Subsequently, this paper focuses on optimizing a mission involving multiple mobile robots charged with inspecting an industrial zone. The aim of this research is to efficiently collect measurements from diverse positions using a fleet of sensing robots operating from a central depot, that is, developing an algorithm for robot decision-making that optimizes mission planning by minimizing an objective function. Initially, we explore our previous proposed solutions and we improve the system by integrating a navigational layer to manage collision avoidance between robots. Then, we delve into scenarios involving multiple homogeneous tasks distributed in a limited geographical environment. To demonstrate feasibility, extensive simulations, numerical experiments, and comparative analysis are conducted, showing the efficiency of the proposed approaches in terms of solution quality and computational complexity. [ABSTRACT FROM AUTHOR]

For a long time, image inspection has been used for surface defect inspection of memory modules. However, its inspection accuracy still does not meet the requirements of mass production, especially for defect inspection of small parts. Deep learning algorithms can improve the inspection accuracy, but they require a considerable amount of actual production defect data. Therefore, in this study, real data for 70,000 pieces of memory modules were obtained during mass production to explore data pre-processing and data augmentation that meets the algorithms' training needs. Due to limited data availability, it is necessary to collect an appropriate number of images and mark the fixed areas or features that represent the more frequently occurring defects in images. The software can learn by itself, speed up the operation, and correctly find the real defect position, effectively improving the detection accuracy so that the algorithm has the advantages of easy training and fast detection speed. The YOLOv5 algorithm has a better detection speed for smaller objects and uses the algorithm's architectural characteristics to flexibly configure models with different complexities, thereby accelerating the convergence as well as simplifying the model architecture and accelerating the calculation speed. During the verification process, the average detection accuracy of production defects can reach 97.5%. It only takes an average of 0.5 s to detect each memory module picture, the yield rate of the production line per quarter is improved by up to 0.08%, and the false positive rate is reduced by 0.12%. In addition, it can improve the efficiency of personnel operations by nearly 40% and save up to 10,000 US dollars in production and operating costs per month. [ABSTRACT FROM AUTHOR]

Defect classification and detection have been explored using convolutional neural networks (CNNs). Normally, a large set of training images containing defects and the associated annotation data are required by these approaches. However, such a large set of images is usually difficult to collect because defects are rare and annotation is time-consuming and expensive. To address these issues, we propose to use a multitask deep one-class CNN for defect classification. Compared with supervised classification methods, this CNN does not require abnormal images and annotated data for training. Specifically, we build a stacked encoder–decoder autoencoder for learning feature representation from normal images. The encoder is used as a feature extractor based on the hard sharing scheme of multitask learning. A one-class classification (OCC) objective learned as a hypersphere using minimum volume estimation is appended to it. Together the encoder and the OCC objective lead to a deep one-class classifier. To train both the autoencoder and one-class classifier end-to-end, a multitask loss function is built. Given an unknown sample, the distance between its feature representation and the center of the hypersphere is used as the anomaly score. Furthermore, defect detection is implemented using a moving-window scanning method on top of the deep one-class classifier. The proposed approach achieves better performance than its counterparts trained using a two-stage method. For defect detection, our approach achieves results almost as good as the supervised method even without using any annotated data. We attribute the promising results to the advantages of multitask learning. Note to Practitioners—Building and evaluating vision-based nondestructive testing (NDT) techniques require many examples of abnormal images, which may not be easy to acquire. This article describes a method that does not require abnormal images for training a convolutional neural network (CNN) in order to perform one-class defect classification (outlier detection). We also applied the method to defect detection with promising results. We include results of experiments demonstrating that better performance can be obtained using our method compared to a set of baselines. Although the proposed method does not use abnormal images for training, it still produces results that are almost as good as the supervised learning-based CNN approaches. This study provides a solution to the challenge encountered by the industrial inspection community when enough abnormal samples are hard to obtain. [ABSTRACT FROM AUTHOR]

High-speed and high-reliability automated quality inspection is widely demanded in the electronics industry. Recently, deep learning theory combined with many non-destructive technologies shows superior performance for inspecting failed soldering. In this paper, a reliable point cloud learning based method is implemented for high-speed solder joint shape defect detection. First, a light-weight neural network named Solder PointNet (SPNet) is proposed. With local group attention mechanisms, SPNet avoids adverse effects of outliers in the scanning point cloud and finds favorable critical point feature adaptively. Then, a modulated loss is designed to ensure reliable low false detection rate. By adjusting the weights of cross-entropy loss, the predictive distribution of defective samples is guided to a smaller range, thereby setting an appropriate threshold to efficiently separate the predictive distribution. Furthermore, the proposed method is further trained and evaluated on the self built solder joint dataset DHU-PAD1000 of point cloud data. Comparison experiments are carried out on the built dataset, and the results show that SPNet achieves higher accuracy with fewer parameters, considerable speed advantage, and more reliable automated detection. [ABSTRACT FROM AUTHOR]

Machine Vision enhanced by Artificial Intelligence can improve product and process monitoring through inspection. However, applying these solutions to the industrial environment requires a transdisciplinary effort and requisites interoperability. This study proposes a road map to reduce implementation uncertainties. A systematic review of the literature was conducted, considering Machine Vision applications with Artificial Intelligence techniques for Industrial Inspection. Natural Language Processing was used to select the potential documents, by narrowing down those that were not aligned with the desired topic. The remaining documents were analyzed and compared with the proposed road map. This allowed the validation of the proposed concepts confirming their applicability. The results show that the road map is valid for current industrial inspections using Artificial Intelligence and Machine vision providing the support for the development of smarter machines into Industry 4.0 functionalities. [ABSTRACT FROM AUTHOR]

In the aerospace manufacturing industry, nondestructive evaluation (NDE) of components plays an important role. Porosities and other defects usually occur in the welds of these components. If such defects end up in the aircraft, the fatigue life of components is lessened, which may cause disastrous accidents. At present, those welds are manually evaluated by human inspectors via reviewing X-ray images. To reduce the workload of inspectors, we have developed an automatic inspection system for identifying defects in linear thin welds. For an X-ray image, this system starts with localizing the central line of the weld using a random forest (RF) regressor. A region surrounding the line is then investigated using an RF classifier in order to detect defects. After extensive experiments, the results demonstrate that the weld can be precisely localized from X-ray images, and the defect detection module can find 80% of defects that have been identified by human inspectors (i.e., true positives), while fewer than 1.6 false positives per image are returned. It is suggested that the system may be beneficial to human inspectors by reducing their workload. In addition, our system produces encouraging results on the publicly available weld X-ray image data set and a magnetic tile image data set. Note to Practitioners—This work was motivated by the challenge of inspecting aerospace components, which is almost entirely done manually at present. Rather than replacing human inspectors, this work aims at reducing their workload by providing them with an initial inspection result for each component. Especially, the proposed system is able to first localize the Region of Interest (RoI) from an X-ray image of a component and then identify potential defects contained in the RoI. To the best of our knowledge, few existing studies perform defect detection on raw component images. Normally, researchers manually cropped an RoI from these images. The output of our system is the pixelwise location information on potential defects. Our results demonstrate that the proposed system is able to accurately localize the weld and identify 80% of defects contained in abnormal weld images with very few false positives. Given that large weld images ($2304\times1920$ pixels) were processed, our system located the weld in 6.6± 1.3 s/image and fulfilled defect detection on each localized weld region in 0.8± 0.1 s. The proposed system was also tested with the publicly available X-ray weld image data set: GDXray and a magnetic tile image data set. Although only a small number of training images were available, promising results were obtained. This suggests that our system is suitable for both X-ray weld images and other images though more work is needed to reduce false positives. [ABSTRACT FROM AUTHOR]

In this paper, an automated inspection system is proposed for detecting the location and measuring the size of existing dipping or buckling on the sandwich panel surface using an RGB camera and an inexpensive linear laser. The proposed method, by observing the radiated laser beams on the sandwich panel surface, can localize and calculate the level of dipping and buckling with acceptable accuracy while being robust to vibrations of moving sandwich panel on the production line conveyor. After a complete processing of the panel by the system, a three-dimensional (surface plot) or two-dimensional (heat map) output is produced to assist the production line supervisor to easily inspect the surface quality of the sandwich panels. Our experimental results show that the proposed system can detect and measure the surface defects including dipping and buckling with a high accuracy and performance. [ABSTRACT FROM AUTHOR]

A new application of muon radiography (MR) is presented in the context of non-destructive testing of industrial equipment. The long-term operation of industrial facilities frequently involves the deterioration of critical components such as pipes and cauldrons due to corrosion and other processes. The precise determination of the inner state of this equipment is needed to ensure the integrity of the facility. MR can be used to infer the thickness of these components through the comparison and further classification of muon observables with respect to well-known templates. A simulation example is presented where the thickness of a pipe made of steel is studied using the Point of Closest Approach method and simple Kolmogorov–Smirnov statistical tests. A precision of about 2–4 mm is obtained using a simple detector with a spatial resolution of 4 mm and exposure times of about 2 h [ABSTRACT FROM AUTHOR]

This work addresses the problem of semi-automatic inspection and navigation in confined environments. A system that overcomes many challenges at the state of the art is presented. It comprises a multirotor able to inspect an industrial combustion chamber thus working in a GPS-denied environment with poor lighting conditions, in the presence of magnetic and communication disturbances, iron dust and repetitive patterns on the structure walls. The presented system is able to pass through narrow entrances but still capable of acquiring high resolution images and to allow operators to perform inspection of the structures. Starting from the captured data, the system is able to provide a 3D reconstruction of the inspected environment for offline analysis. [ABSTRACT FROM AUTHOR]

A simple dichroic splitter for dual-band line scanning is described. It comprises prisms elements that enable cheapness of the whole prototype by using only one linear detector. Validity of the design is demonstrated via in-line moisture measurement. [ABSTRACT FROM AUTHOR]

Automating thermographic analysis for industrial and construction inspection is one of the essential aspects of NDT 4.0 in industrial transformation. Thermal segmentation is a preliminary step for characterizing a component's heat signatures by detecting and extracting thermal patterns. Manual segmentation of areas with distinctive thermal patterns is a time-consuming task that consequently can increase the operation cost. Thus, automatic thermal segmentation can significantly assist companies in avoiding possible misinterpretation and decrease the operation cost. Due to the lack of enough labeled thermal images to train deep neural networks and the fact that, by nature, thermal abnormalities may have non-uniform patterns and form in different shapes and sizes, training a deep neural network can be a challenging task. Moreover, selecting common features can cause possible convergence onto dependency toward specific shapes or patterns. Thus, this paper proposes a self-training thermal segmentation method involving a novel process pipeline for using labeled and unlabeled data for training the presented deep-learning model. The method uses an unsupervised approach for predicting the ground truth in case of unlabeled data before passing the data to the secondary model for segmentation. For this paper, three experiments were conducted to evaluate the method's performance and provide a comparative analysis with the most recent techniques. The first experiment used a simulated inspection created by COMSOL software. The second experiment employed several plate inspections in several isolated environments. Finally, the third experiment used thermal images acquired during an actual inspection of the piping structure and aerospace components. [Display omitted] • Propose a self-training thermal segmentation method for the analysis of thermal images. • Introduce an unsupervised subnet for labeling unlabeled data. • Present a dataset of simulated plate inspection conducted using active thermography. [ABSTRACT FROM AUTHOR]

An inspection system using estimated three-dimensional (3-D) surface characteristics information to detect and classify the faults to increase the quality control on the frequently used industrial components is proposed. Shape from shading (SFS) is one of the basic and classic 3-D shape recovery problems in computer vision. In our application, we developed a system using Frankot and Chellappa SFS method based on the minimization of the selected basis function. First, the specialized image acquisition system captured the images of the component. To eliminate noise, wavelet transform is applied to the taken images. Then, estimated gradients were used to obtain depth and surface profiles. Depth information was used to determine and classify the surface defects. Also, a comparison made with some linearization-based SFS algorithms was discussed. The developed system was applied to real products and the results indicated that using SFS approaches is useful and various types of defects can easily be detected in a short period of time. [ABSTRACT FROM AUTHOR]

• Propose a new DAVOLOv5 model based on domain adaptation. • Add a transfer term to the loss function. • Use model ensemble methods for hard-to-detect samples. • Recommend a hyperparameter controlling the domain representations. In the process of industrial production, manufactured products are prone to surface defects for a variety of reasons. To overcome the problem of high time cost and the strong demand for large sample data sets, a detector based on transfer learning is commonly utilized. In this paper, a domain adaptation YOLOv5 model, named DAYOLOv5, is proposed for automatic surface defect inspection. The hyperparameter α in DAYOLOv5 for knowledge transfer can be designed specially to achieve better generalization in real-world industrial applications. Meanwhile, in the field of magnetic tile surface defect detection, our DAYOLOv5 outperforms traditional mixed training and pretrain-finetune methods with limited data sets and has great robustness. Overall, the experimental results demonstrate that our DAYOLOv5 model can indeed boost performance on small-scale target data sets and is applicable to practical industrial scenarios. [ABSTRACT FROM AUTHOR]

The acquisition of three-dimensional surface data plays an increasingly important role in the industrial sector. Numerous 3D shape measurement techniques have been developed. However, there are still limitations and challenges in fast measurement of large-scale objects or high-speed moving objects. The innovative line scan technology opens up new potentialities owing to the ultra-high resolution and line rate. To this end, a sensor for in-motion continuous 3D shape measurement based on dual line-scan cameras is presented. In this paper, the principle and structure of the sensor are investigated. The image matching strategy is addressed and the matching error is analyzed. The sensor has been verified by experiments and high-quality results are obtained. [ABSTRACT FROM AUTHOR]

An application of spectral domain optical coherence tomography (SD-OCT) was demonstrated for a fast industrial inspection of an optical thin film panel. An optical thin film sample similar to a liquid crystal display (LCD) panel was examined. Two identical SD-OCT systems were utilized for parallel scanning of a complete sample in half time. Dual OCT inspection heads were utilized for transverse (fast) scanning, while a stable linear motorized translational stage was used for lateral (slow) scanning. The cross-sectional and volumetric images of an optical thin film sample were acquired to detect the defects in glass and other layers that are difficult to observe using visual inspection methods. The rapid inspection enabled by this setup led to the early detection of product defects on the manufacturing line, resulting in a significant improvement in the quality assurance of industrial products. [ABSTRACT FROM AUTHOR]

The interest in gonioapparent pigments (metallic, pearlescent, interference, or diffractive) has increased in the last few years, especially for applications in the automotive industry. To assure a proper characterization of colors with gonioapparent pigments, commercial devices have appeared to characterize the color in different geometries, which are called multiangle spectrophotometers. As the gonioapparent pigments and multiangle instruments are relatively new, no studies exist regarding the instrumental-based procedure followed in the industry, and if the results provided are in agreement with the observer perception. Consequently, the main objective of this study was to examine the correlation of the instrumental color differences with visual assessments. The instrumental color difference was calculated with the color difference formula AUDI2000 (specific for this sector) between the pairs of similar samples of three types of coated panels (solid, metallic, and pearlescent). The values measured by a telespectroradiometer in a directional lighting booth and the colorimetric values obtained by means of a multiangle spectrophotometer BYK-mac were considered for this purpose. Additionally, a visual experiment was conducted to quantify the color difference by using the gray-scale method. The results revealed that an acceptable instrumental correlation existed despite the visual and the instrumental correlation being worse. In particular, it was checked that observers accepted a larger number of color pairs, that is, the visual color difference was smaller than the tolerance demanded by the industry (derived from AUDI2000). © 2015 Wiley Periodicals, Inc. Col Res Appl, 41, 384-391, 2016 [ABSTRACT FROM AUTHOR]

Steel rails used in the transport sector and in industry are designed and manufactured to support high stress levels generated by high-speed and heavy-loaded modern trains. In the rail manufacturing process, one of the key stages is rolling, where fast, accurate, and repeatable rail profile measurement is a major challenge. In this paper, a rail profile measurement system (PMS) for rail rolling mills based on four conventional, inexpensive laser range finders is proposed. The range finders are calibrated using a common reference to properly express the point clouds generated by each range finder in the world coordinate system (WCS). The alignment of the point clouds to the rail model is performed by means of an efficient and robust registration method. Experiments carried out in a rail rolling mill demonstrate the accuracy and repeatability of the system; the maximum error is below 0.12%. All parallelizable tasks were designed and developed to be executed concurrently, achieving an acquisition rate of up to 210 fps. [ABSTRACT FROM AUTHOR]

Laser triangulation is a well-established technique in 3D surface metrology. However, scattering surfaces and reflectivity variations cause measurement uncertainties due to a reduced signal-to-noise ratio. To improve the measurement accuracy of such surfaces a new laser line detection algorithm is proposed. Within this work, the laser line segmentation and, therefore, the separation of the laser line and the noisy background in the camera images are based on the Chan–Vese model. The Chan–Vese model is adapted to reduce the computation time and increase the measurement rate, which is important in industrial applications. In this paper, complete instructions on how to apply the Chan–Vese segmentation algorithm to laser line triangulation measurements are given, including initialization and a parameter set for the segmentation process. Further, an example of laser triangulation measurement of a microstructured, highly scattering surface is presented. Closing, the proposed approach is compared with a conventional line detection method. [ABSTRACT FROM AUTHOR]

Noncontact optical imaging is frequently used in the inspection and metrology of stationary objects, including in particular the reconstruction of the 3-D surface profile. A technique, known as phase-measuring profilometry, involves projecting a sinusoidal pattern and then inferring the height of various points on the object by measuring the resulting phase changes at the respective locations. However, this method cannot be directly applied to systems involving moving objects, as the translation and the perspective geometry effect manifest as errors in the height calculations. In this paper, we report on an imaging and numerical surface profilometry with error compensation technology (INSPECT) measurement system that is tailored for moving objects. We model the imaging system that considers the nonlinear perspective geometry effect, and simplify to a first-order equation using Taylor series expansion. With this, we generalize the conventional phase shift algorithm, and develop the optimization procedures that can compute the height information effectively. We apply this technology to the INSPECT measurement system in semiconductor manufacturing and show significant improvement in accuracy and robustness. [ABSTRACT FROM PUBLISHER]

We present a multi-line-scan light-field image acquisition and processing system designed for 2.5/3-D inspection of fine surface structures in industrial environments. The acquired three-dimensional light field is composed of multiple observations of an object viewed from different angles. The acquisition system consists of an area-scan camera that allows for a small number of sensor lines to be extracted at high frame rates, and a mechanism for transporting an inspected object at a constant speed and direction. During acquisition, an object is moved orthogonally to the camera's optical axis as well as the orientation of the sensor lines and a predefined subset of lines is read out from the sensor at each time step. This allows for the construction of so-called epipolar plane images (EPIs) and subsequent EPI-based depth estimation. We compare several approaches based on testing a set of slope hypotheses in the EPI domain. Hypotheses are derived from block matching, namely the sum of absolute differences, modified sum of absolute differences, normalized cross correlation, census transform, and modified census transform. Results for depth estimation and all-in-focus image generation are presented for synthetic and real data. [ABSTRACT FROM AUTHOR]

This paper presents an alternative method of material damage evaluation based on the X-ray computer tomography-detected microdefects and multiscale computer simulation. This is achieved by developing a method of the digital diagnosis and full-field numerical calculation of material degradation in macroscopic material test specimens. The method comprised three basic components: (a) digital detection and processing of micro/mesoscale material defects of macroscopic material test specimens; (b) multilevel meshing and multilevel finite element analysis for evaluating local/global material degradation; and (c) synchronized experimental and numerical determination of material damage. The unique contributions of the proposed approach include (a) a multilevel finite element meshing and analysis scheme that makes the full-field estimation of material degradation in macroscopic test specimens computationally tractable on regular workstations, (b) full-field exploration of mesoscale material defects (i.e., those with a feature size from several micrometers to a few millimeters), which play a crucial role in failure analysis of engineering components, and (c) the proposed method offers a significantly better accuracy in estimating material degradation in terms of effective modulus than the conventional analytical models in continuum damage mechanics and micromechanics. Test results of aluminum alloys confirm the efficacy of our approach in the digital interrogation of material degradation. [ABSTRACT FROM PUBLISHER]

We address the methodology and hardware design for detecting defects on a steel surface. The goal of this work is to develop a surface inspection system for defects with three-dimensional (3-D) characteristics on planar steel surface using multispectral photometric stereo technique. First, the reflections of multispectral illuminations from a steel surface are captured by a multi-channel camera system. Then, a 3-D reconstructed surface profile is extracted from estimated gradient fields. The developed system is applied to real steel specimens and the experimental results show that it can reliably detect tiny defects on the surface. [ABSTRACT FROM AUTHOR]

The variety of vision inspection systems for welding defects in the present manufacturing scenario is needed for overcoming certain limitations such as the problem of inaccuracy in the images, non-uniformed illumination, noise and deficient contrast, and confusion in defects if they occur in the same spot at the surface and subsurface. Hence, it is imperative to design a new vision inspection system which will enable to overcome the aforementioned problems in welding. A sophisticated new vision inspection system using machine vision has been developed for this study to identify and classify the surface defects of butt joint as per standard EN25817 in metal inert gas (MIG) welding. In this proposed vision system, images of welding surfaces are captured through a CCD camera. Four frames of sequence of images are obtained using four zones of LEDs using the front light illumination system in this method. From these images, the regions of interest are segmented and the average gray levels of the characteristic features of these images are calculated. The same process can be extended further to four zones (four quadrants) of four types of welded joints. Finally, welded joints can be classified into one of the four predefined ones based on the back-propagation neural network. The proposed system demonstrates an overall accuracy of 95% from the 80 real samples tested. [ABSTRACT FROM AUTHOR]

On-line non-contact surface inspection with high precision is still an open problem. Laser triangulation techniques are the most common solution for this kind of systems, but there exist fundamental limitations to their applicability when high precisions, long standoffs or large apertures are needed, and when there are difficult operating conditions. Other methods are, in general, not applicable in hostile environments or inadequate for on-line measurement. In this paper we review the latest research in Conoscopic Holography, an interferometric technique that has been applied successfully in this kind of applications, ranging from submicrometric roughness measurements, to long standoff sensors for surface defect detection in steel at high temperatures. [ABSTRACT FROM AUTHOR]

Object detection with high accuracy is becoming increasingly important for industrial processes, such as producing printed circuit boards (PCBs). A false or missed detection during production can lead to serious quality issues. Therefore, an efficient detector that can maintain high quality is required for industrial applications. This paper proposes a method for improving detection accuracy while supporting real-time operations using the baseline YOLOv4. A new loss function for box regression called Gaussian intersection of union (GsIoU) is explored, which merges the predicted boxes under different anchors at the same position using a Gaussian function to calculate the box regression loss, improving the accuracy of the final box regression. The proposed PCB Component (PCBC) dataset is a benchmark comprising 18,948 images, 24 categories (containing the same component in different directions), and 508,313 components. Taking YOLOv4 as the experimental baseline on the PCBC dataset, the mean average precision (mAP) using the GsIoU has reached 86.9%, which is improved by 3.3% and 2.2% compared to using the CIoU and GIoU loss functions, respectively. The experiments were conducted on the COCO dataset to verify the generalization of the GsIoU. The detection accuracy of the GsIoU exceeds that GIoU and CIoU by 0.6% on AP 50 and 0.6% on AP 50:95 , achieving 65.6% and 46.0% on the COCO test-val2017, respectively. The detection efficiency of the proposed method is the same as that of the baseline in the testing process and is slightly reduced in the training process owing to the synchronous calculation of the IoU, variance, and Gaussian operations on the network output. The experiments indicate that GsIoU is effective and efficient. • A novel Gaussian intersection over union is designed for box regression. • An adaptive weight strategy is introduced for loss function. • A new dataset of circuit board components (PCBC dataset) is created. • The proposed method achieves good performance on COCO dataset and PCBC dataset. [ABSTRACT FROM AUTHOR]

Advances in electronic technology have made integrated circuits (ICs) the fundamental components in all electronic devices. To increase their production yield by catching defects as early as possible, we need to perform quality assurance on the semiconductor dies during the assembly and packaging processes. A common approach is to employ machine vision to compare a test die with a "known good die". However, difficulties in ensuring identical imaging conditions (such as illumination) are limitations to this die-to-die comparison approach. Instead, in this work we develop a novel reference-free defect detection algorithm for an IC die by analyzing its image. By identifying intrinsic and extrinsic features of various segments in the image, we implement a classification scheme to identify whether the die is defective or not. We rely on the fact that normal ICs contain regular patterns, and the abnormal and irregular regions are classified as potential areas of defects. Experimental results show that the proposed reference-free defect detection algorithm can detect most of the defects from different types of IC dies, and can also correctly classify normal IC dies as non-defective. These results demonstrate the feasibility of the reference-free defect detection approach. [ABSTRACT FROM AUTHOR]

The variable visual acuity nature of human vision across fovea and periphery prompted the need to study the differential magnification method as compared with the traditional linear magnification. With differential magnification, outer-area objects were magnified at a level higher than the inner-area objects. An application software was specially developed by the authors previously for the examination of differential linear and traditional linear magnification methods. This paper presents the enhancement of the application software and the addition of other magnification interface factors which worth further investigating. The newly designed differential nonlinear magnification method was featured and comparisons were made with the differential linear and the traditionally linear magnification methods. [ABSTRACT FROM AUTHOR]

The investment in an automated fabric defect detection system is more than economical when reduction in labor cost and associated benefits are considered. The development of a fully automated web inspection system requires robust and efficient fabric defect detection algorithms. The inspection of real fabric defects is particularly challenging due to the large number of fabric defect classes, which are characterized by their vagueness and ambiguity. Numerous techniques have been developed to detect fabric defects and the purpose of this paper is to categorize and/or describe these algorithms. This paper attempts to present the first survey on fabric defect detection techniques presented in about 160 references. Categorization of fabric defect detection techniques is useful in evaluating the qualities of identified features. The characterization of real fabric surfaces using their structure and primitive set has not yet been successful. Therefore, on the basis of the nature of features from the fabric surfaces, the pro- posed approaches have been characterized into three categories; statistical, spectral and model-based. In order to evaluate the state-of-the-art, the limitations of several promising techniques are identified and performances are analyzed in the context of their demonstrated results and intended application. The conclusions from this paper also suggest that the combination of statistical, spectral and model-based approaches can give better results than any single approach, and is suggested for further research. [ABSTRACT FROM AUTHOR]

In this paper, we present a robust image alignment algorithm based on matching of relative gradient maps. This algorithm consists of two stages; namely, a learning-based approximate pattern search and an iterative energy-minimization procedure for matching relative image gradient. The first stage finds some candidate poses of the pattern from the image through a fast nearest-neighbor search of the best match of the relative gradient features computed from training database of feature vectors, which are obtained from the synthesis of the geometrically transformed template image with the transformation parameters uniformly sampled from a given transformation parameter space. Subsequently, the candidate poses are further verified and refined by matching the relative gradient images through an iterative energy-minimization procedure. This approach based on the matching of relative gradients is robust against nonuniform illumination variations. Experimental results on both simulated and real images are shown to demonstrate superior efficiency and robustness of the proposed algorithm over the conventional normalized correlation method. [ABSTRACT FROM AUTHOR]

A random visual search model was fit to 2980 search times obtained from a double-target search task. The overall adequacy and prediction accuracy of the model for various search time parameters like mean and median search times, and response times for both individual and pooled data show that subjects' search strategy may reasonably be inferred from search time distributions. The results also suggest the general applicability of the random search model for describing the search behaviour of a large number of subjects performing the type of search task used here, and the practical feasibility of its application for determination of stopping policy in optimization of inspection system design. Notwithstanding the general conformity of the data with the model, some extremely long search times were found, and this was accounted for by the characteristics of visual lobes, nonoptimum interfixation distances and inappropriate overlapping of lobes previously reported. An interpretation of temporal changes of subjects' search strategies during target detection was made by considering the cross-over phenomenon exhibited by the two cumulative detection percentage curves. [ABSTRACT FROM AUTHOR]

• The optimal band selection is applied to design a network module. • A method is proposed for evaluating and optimizing the convolutional kernel. • The optimal kernels and features show semantics and make the network robust. • The impulsiveness or cyclostationarity can be used as semantics. • An improvement of 4% and 3% is achieved under two scenarios, respectively. Deep data-driven methods for fault diagnosis, as an engineering-oriented approach, rely heavily on target data. For engineering applications, the working conditions of rotating machinery fluctuate from time to time and a collection for any working conditions is impossible. To tackle this problem, a robust network with optimal kernels named EnvelopeNet is proposed for extracting solid information and eliminating the influence of fluctuations. In the EnvelopeNet, a feature evaluation building block named envelope module is constructed based on optimal band selection theory to optimize the kernels. Compared with the kurtogram, the learned optimal kernels and features show strong semantics which helps the network become robust. The EnvelopeNet is validated under the approximate speed and mixed speed scenarios. The results show that the EnvelopeNet could provide admirable generalization ability for fluctuating working conditions and an average improvement of about 4% and 3% over existing approaches under two scenarios respectively. [ABSTRACT FROM AUTHOR]

Automatic industrial inspection is critical to modern manufacturing enterprises. Due to their low cost and real-time processing speed, vision-based inspection systems are often used for this task. Deep Convolutional Neural Networks (CNNs) have been extensively applied to many computer vision tasks but require large numbers of annotated examples to train. Obtaining such annotations is expensive and time-consuming. In this study we describe a method which aims to make the best use of unannotated image data, which can often be collected easily. We propose a novel unsupervised local deep feature learning method based on image segmentation to build a network which can extract useful features from an image. The training algorithm alternates between (1) obtaining pseudo-labels by clustering the features extracted using a segmentation CNN and (2) training the CNN for feature learning using these pseudo-labels. To our knowledge, unsupervised local deep feature learning has not been addressed based on image segmentation in this way before. We demonstrate the approach on two aerospace weld inspection tasks. Our results show that the proposed unsupervised method performs almost as well as a CNN with the same architecture trained in a supervised manner. • Introduction of an unsupervised local deep feature learning method. • Developing an automatic aerospace weld inspection system on top of the proposed method. • Comparing the proposed method with the unsupervised classification CNN with the same backbone. [ABSTRACT FROM AUTHOR]

Nowadays, unmanned aerial vehicles (UAVs) are extensively used for multiple purposes, such as infrastructure inspections or surveillance. This paper presents a real-time path planning algorithm in indoor environments designed to perform contact inspection tasks using UAVs. The only input used by this algorithm is the point cloud of the building where the UAV is going to navigate. The algorithm is divided into two main parts. The first one is the pre-processing algorithm that processes the point cloud, segmenting it into rooms and discretizing each room. The second part is the path planning algorithm that has to be executed in real time. In this way, all the computational load is in the first step, which is pre-processed, making the path calculation algorithm faster. The method has been tested in different buildings, measuring the execution time for different paths calculations. As can be seen in the results section, the developed algorithm is able to calculate a new path in 8–9 milliseconds. The developed algorithm fulfils the execution time restrictions, and it has proven to be reliable for route calculation. [ABSTRACT FROM AUTHOR]

• Cylinder primitive extraction is addressed as a clustering problem of possible cylinder candidates. • Candidates are generated by cutting the scene through several random planes. • A novel cylinder similarity function based on dual quaternion algebra is defined. • The clustering process, based on Game Theory, inherently discards incompatible candidates. • The proposed method can deal with noisy, partially occluded and non-oriented point clouds. Finding geometric primitives in 3D point clouds is a fundamental task in many engineering applications such as robotics, autonomous-vehicles and automated industrial inspection. Among all solid shapes, cylinders are frequently found in a variety of scenes, comprising natural or man-made objects. Despite their ubiquitous presence, automated extraction and fitting can become challenging if performed "in-the-wild", when the number of primitives is unknown or the point cloud is noisy and not oriented. In this paper we pose the problem of extracting multiple cylinders in a scene by means of a Game-Theoretic inlier selection process exploiting the geometrical relations between pairs of axis candidates. First, we formulate the similarity between two possible cylinders considering the rigid motion aligning the two axes to the same line. This motion is represented with a unitary dual-quaternion so that the distance between two cylinders is induced by the length of the shortest geodesic path in SE(3). Then, a Game-Theoretical process exploits such similarity function to extract sets of primitives maximizing their inner mutual consensus. The outcome of the evolutionary process consists in a probability distribution over the sets of candidates (ie axes), which in turn is used to directly estimate the final cylinder parameters. An extensive experimental section shows that the proposed algorithm offers a high resilience to noise, since the process inherently discards inconsistent data. Compared to other methods, it does not need point normals and does not require a fine tuning of multiple parameters. [ABSTRACT FROM AUTHOR]

This paper expands upon a previous publication and is the natural continuation of an earlier study which presented an industrial validator of expiration codes printed on aluminium or tin cans, called MONICOD. MONICOD is distinguished by its high operating speed, running at 200 frames per second and validating up to 35 cans per second. This paper adds further detail to this description by describing the final stage of the MONICOD industrial validator: the process of effectively validating the characters. In this process we compare the acquired shapes, segmented during the prior stages, with expected character shapes. To do this, we use a template matching scheme (here called "morphologies") based on bitwise operations. Two learning algorithms for building the valid morphology databases are also presented. The results of the study presented here show that in the acquisition of 9885 frames containing 465 cans to be validated, there was only one false positive (0.21% of the total). Another notable feature is that it is at least 20% faster in validation time with error rates similar to those of classifiers such as support vector machines (SVM), radial base functions (RBF), multi-layer perceptron with backpropagation (MLP) and k-nearest neighbours (KNN). [ABSTRACT FROM AUTHOR]