Search

Catastrophic optical damage (COD) is one of the reasons limiting the output power and lifetime of semiconductor lasers. Nevertheless, the automatic defects inspection of the COD is a challenging task due to several factors, including the micron-scale size of the laser chip, poor contrast, the notable similarities in defect features across different categories, and the fact that defects occupy a minimal portion of image pixels in the images. In this work, We first design and implement a data acquisition and inspection system to collect micron-scale laser chip electroluminescence (EL) images. Secondly, we establish a laser chip COD dataset for training. Finally, a novel COD detection network (CODDNet) is proposed to construct an end-to-end defect detection method. To better extract strip-like COD features under poor contrast, a strip convolution is proposed to acquire more discriminative features and reduce the parameters. A multi-scale strip convolution aggregation structure is proposed to extract richer information across different defect categories from the network to obtain multi-scale feature maps. To address the class imbalance issue between defect and the background, an attention module is embedded into the block to emphasize the COD defects. The experimental results demonstrated that the proposed CODDNet could achieve a higher inspection accuracy and faster inference speed with fewer parameters. Based on the proposed method, manufacturers could take corresponding effective measures to improve the stability of laser chips. In the future, we will continue to expand the COD dataset and conduct research on defect detection for different types of laser chips. [ABSTRACT FROM AUTHOR]

The use of machine vision and deep learning for intelligent industrial inspection has become increasingly important in automating the production processes. Despite the fact that machine vision approaches are used for industrial inspection, deep learning-based defect segmentation has not been widely studied. While state-of-the-art segmentation methods are often tuned for a specific purpose, extending them to unknown sets or other datasets, such as defect segmentation datasets, require further analysis. In addition, recent contributions and improvements in image segmentation methods have not been extensively investigated for defect segmentation. To address these problems, we conducted a comparative experimental study on several recent state-of-the-art deep learning-based segmentation methods for steel surface defect segmentation and evaluated them on the basis of segmentation performance, processing time, and computational complexity using two public datasets, NEU-Seg and Severstal Steel Defect Detection (SSDD). In addition we proposed and trained a hybrid transformer-based encoder with CNN-based decoder head and achieved state-of-the-art results, a Dice score of 95.22% (NEU-Seg) and 95.55% (SSDD). [ABSTRACT FROM AUTHOR]

Dark-line defects in the active region of vertical-cavity surface-emitting lasers (VCSELs) reduce optical power and compromise system stability. Therefore, it is of paramount importance to identify these defects, which can help improve production processes. Traditional methods rely on visual inspection or electrical characteristics, requiring specialized knowledge. We propose an end-to-end defect segmentation network with a novel dual-encoder mechanism to extract both local and global features of defects. In response to the challenges posed by random sizes and varying shapes of defects, we propose a dynamic convolution encoder (DCE) to extract local features. To capture the overall distribution of defects, we present the Shifted MLP Encoder (SME) to extract global context information. A hybrid loss function is employed to improve the ability to identify defect samples to address the imbalance between positive and negative samples. Additionally, deep supervision is applied to oversee multi-scale features in intermediate layers, enhancing segmentation performance. Finally, experimental results on the VCSEL dataset we made demonstrate that the dice similarity coefficient (DSC) and mean intersection over union (mIoU) of defect segmentation reach 90.72% and 91.58%, respectively. Compared with other advanced methods, our proposed method improves segmentation accuracy with only a slight increase in computational complexity. [ABSTRACT FROM AUTHOR]

In the paper-based packaging field, a sheet of paper material is introduced in a die cutter to create several creases and cuts. These creases allow to fold the sheet, and consequently, to build the final package. Thus, the quality of the crease is critical to get a valuable final product. In this work, a new system for automatic crease inspection is presented. The inspection is performed by processing crease visual information captured by a laser sensor. Automation is achieved by moving the laser with a cartesian robot along the sheet. A crease profile processing algorithm is proposed as well, which is able to extract geometric features from the different patterns that are associated with different paper types. Moreover, a crease classification methodology is proposed, where creases are grouped taking into account some physical properties. The experiments performed show the viability and usefulness of the proposed system. [ABSTRACT FROM AUTHOR]

Nowadays, the number of aging civil infrastructures is growing world-wide and when concrete is involved, cracking and delamination can occur. Therefore, ensuring the safety and serviceability of existing civil infrastructure and preventing an inadequate level of damage have become some of the major issues in civil engineering field. Routine inspections and maintenance are then required to avoid leaving these defects unexplored and untreated. However, due to the limitations of on-field inspection resources and budget management efficiency, automation technology is needed to develop more effective and pervasive inspection processes. This paper presents a pixel-wise classification method to automatically detect and quantify concrete defects from images through semantic segmentation network. The proposed model uses Deeplabv3+ network with weights initialized from pre-trained neural networks. The comparison study among the performance of different deep neural network models resulted in ResNet-50 as the most suitable network for applications of civil infrastructure defects segmentation. A total of 1250 images have been collected from the Internet, on-field bridge inspections and Google Street View in order to build an invariant network for different resolutions, image qualities and backgrounds. A randomized data augmentation allowed to double the database and assign 2000 images for training and 500 images for validation. The experimental results show global accuracies for training and validation of 93.42% and 91.04%, respectively. The promising results highlighted the suitability of the model to be integrated in digitalized management system to increase the productivity of management agencies involved in civil infrastructure inspections and digital transformation. [ABSTRACT FROM AUTHOR]

Site inspection is a notably tedious, time-consuming, and error-prone process when carried out manually by construction inspectors. One portion of site inspection is the generation and comparison of as-built drawings to their as-planned counterparts to ensure conformity. Current as-built evaluation systems rely on manually positioning and transporting scanning stations around the site to collect a cohesive map of the environment and then followed by a manual assessment of the site status using the built drawings. This paper proposes an automated alternative to the generation and assessment of 2D as-built floor maps by relying on robotic simultaneous localisation and mapping (SLAM) to build the 2D as-built maps, followed by automated evaluation of the as-built maps using machine vision. The components of the proposed system have been tested through proof of concept experiments inside 2 different constructed sites. Results indicate average errors below 4 cm (site of 30 m x 15 m) between the disparities identified by our proposed system versus ground truth measured disparities. [ABSTRACT FROM AUTHOR]

Industrial X-ray analysis is common in aerospace, automotive or nuclear industries where structural integrity of some parts needs to be guaranteed. However, the interpretation of radiographic images is sometimes difficult and may lead to two experts disagree on defect classification. The automated defect recognition (ADR) system presented herein will reduce the analysis time and will also help reducing the subjective interpretation of the defects while increasing the reliability of the human inspector. Our convolutional neural network (CNN) model achieves 0.942 mAP@IoU = 0.50, which is considered as similar to expected human performance, when applied to an automotive aluminium castings dataset (GDXray). On an industrial environment, its inference time is less than 400 ms per 16 GB DICOM image (16 bits), so it can be installed on production facilities with no impact on delivery time. In addition, an ablation study of the main hyper-parameters to optimise model accuracy from the initial baseline result of 0.75 mAP up to 0.942 mAP, was also conducted. [ABSTRACT FROM AUTHOR]

In this paper, an automated layer defect detection system for construction 3D printing is proposed. Initially, a step-by-step procedure is implemented to develop a deep convolutional neural network that receives images as input and is able to distinguish concrete layers from other surrounding objects through semantic pixel-wise segmentation. Using data augmentation techniques, 1M labeled images are generated and used to train and test the CNN model. Then, a defect detection module is developed which is able to detect deformations in the printed concrete layers extracted from the images using the CNN model. The evaluation results based on metrics such as accuracy, F1 score, and miss rate verify the acceptable performance of the developed system. [ABSTRACT FROM AUTHOR]

The automatic phased array ultrasonic testing method has been developed for inspecting the pipe welds with complex geometry. Applying time of flight diffraction (TOFD) technique has two problems: one is the difficulty in recognition of head wave under complex structure; the other is the limited scanning space cannot provide the needed span for pitch-catch mode. To overcome these two problems, an array probe can be placed on the surface of base metal of the weld by the direct and double traverse techniques. In order to design the optimal scanning path for high probability of detection and inspection rate, the model of beam path on tilt steps was established to quantitatively analyze the effects of tilt steps on beam coverage and mode conversion. Experiments were conducted to validate the proposed rule of planning the optimal scanning path. All artificial defects in the mock-up block were detected on the scanning side of which exists the optimal scanning path. While for the scanning side without the optimal scanning path, a combination of scanning is needed to avoid missing inspection. [ABSTRACT FROM AUTHOR]

Castings produced for the automotive industry are considered important components for overall roadworthiness. To ensure the safety of construction, it is necessary to check every part thoroughly using non-destructive testing. X-ray testing rapidly became the accepted way of controlling the quality of die-cast pieces. In this paper, we present a Convolutional Neural Network (CNN) for defect detection in castings. In order to train the CNN model, a large dataset is necessary. We build the dataset by using synthetic defects. They are simulated using 3D ellipsoidal models and Generative Adversarial Networks (GAN). We compare different portions of ellipsoidal/GAN defects in the training subset. In our experiments, the use of GAN defects does not play a relevant role in this solution. However, ellipsoidal defects helped to achieve better performance. Ellipsoidal defects from any size and orientation could be superimposed onto real X-ray images in any location. In addition, we tested several CNN configurations, the best one, that we call Xnet-II, has 30 layers and more than 1,350,000 parameters. It has been trained using a dataset with around 640,000 patches containing 50% of ellipsoidal defects and 50% of real background captured from different casting types. The model was tested using sliding-windows methodology on whole X-ray images achieving promising results (mPA = 0.7102): the model was able to detect real defects from different casting types. We believe that the methodology presented could be used in similar projects that have to deal with automated detection of defects. [ABSTRACT FROM AUTHOR]

When a silicon wafer is cut into separate dies, their front and back sides might have chipping resulting in die cracks and yield loss. To prevent defect formation, silicon wafers should undergo optical inspection for evaluation of wafer chipping, its size, and its shape. This work proposes an automated method of image processing that includes die edge detection, die street search, and determination of chipping size and shape. Die edge search was done using an Otsu’s thresholding method. This technique was chosen out as the optimal of the five ones. The choice was based on the segmentation precision evaluation of two types of images: with sharp and blurred edges. Die street search was done using a developed algorithm capable of processing images with angular displacement. Chipping shape and size were calculated through die edge displacement from the die street. Based on the numerical evaluation of chipping size and shape, a chipping danger metric that may be used for detection of defective dies has been proposed. [ABSTRACT FROM AUTHOR]

Automated inspection of manufactured parts is increasingly getting attention as it helps to make a rapid decision on product conformity. In this context, the aim of this paper is to present a new scanner path planning method for part inspection using an industrial six-axis robot. The novelty of the approach is to generate a scan path with the control of the overlap between two adjacent scanning paths based on the use of the least-squares conformal maps, which stretches a 3D mesh surface on a 2D plane. Equidistant paths calculated in the 2D space are then transformed into equidistant paths in the 3D space. The effective performance of controlling the overlap can improve digitizing quality and save digitizing time by managing the coverage of the laser beam. Furthermore, the digitizing quality is also ensured by keeping a constant scanning distance and executing a continuous control of the scanner orientation relatively to the part surface for all the driven points of the scan path. An experimental application of this new approach is proposed for a laser-plane scanner mounted on an industrial robot with six degrees of freedom, which demonstrates the interest of such an approach. [ABSTRACT FROM AUTHOR]

There are various surface defects which occur during the hot rolling of steels. It is difficult to correctly identify and control these defects due to the different inspection techniques on different materials and sizes. Also, the statistical data analysis techniques typically used like the principal component analysis, factor analysis etc. require a lot of plant data and are computationally very intensive. Before a detailed analysis of the actual cause of the defects can be done, it is necessary to separate the defects as those coming from the continuous casting or the rolling mill. Once this is done, analysis on the individual components can then be completed to find the root cause. To accomplish both these analysis, Bayesian hierarchical modeling is done on the automated inspection of the bars to form a causal relationship of the defects to the process equipments. Variance reduction model is used at the top of the analysis and regression models are used in the next level. [ABSTRACT FROM AUTHOR]

A vision-based inspection system has been investigated in order to improve the quality of products and processes found in various industries. In this paper, we propose a new defect detection algorithm for steel wire rods produced by the hot rolling process. Because the steel wire rods are long cylinder rods with a circular cross section, the brightness at the sides and center is inconsistent. Moreover, the various types of steel wire rods and the presence of scales affect the reflection properties of the rod surface. In order to resolve the abovementioned difficulties, the use of dynamic programming and a discrete wavelet transform are proposed. An adaptive local binarization method is used to further reduce the effects of scale. The effectiveness of the proposed method is shown by means of experiments conducted on images of steel wire rods that were obtained from an actual steel production line. [ABSTRACT FROM AUTHOR]

In the last few years the need for methodologies capable of performing an automated geometric inspection has increased. These methodologies often use 3D high-resolution optical digitisers to acquire points from the surface of the object to be inspected. It is expected that, in the near future, geometric inspection will be requiring more and more the use of these instruments. At present geometric inspection is not profiting from all the opportunities attainable by 3D high-resolution optical scanners or from the numerous tools which can be used for processing the point cloud acquired from the inspected product. For some years now, these authors have been working on a new methodology for automatic tolerance inspection working from a 3D model acquired by optical digitisers. In this paper all the information recognisable in a scanned object is organised into a new data structure, called Recognised Geometric Model ( RGM). The final aim is to define a representation of the inspected object for the automatic evaluation of the non-idealities pertaining to the form, orientation and location of the non-ideal features of the acquired object. The key concept of the proposed approach is the capability to recognise some intrinsic nominal properties of the acquired model. These properties are assumed as references to evaluate the non-idealities of the inspected object. With this approach the references of geometric inspection are searched for in the inspected object independently of a tolerance specification and of the availability of a 3D nominal representation. The high-level geometric information within RGM depends on the rules used for its identification. The capability to recognise specific categories of nominal references offers the possibility of introducing new tolerances to be specified. The proposed approach has been implemented in original software by means of which a specific test case has been analysed. [ABSTRACT FROM AUTHOR]

One of the industrial applications of computer vision is the automated detection and characterisation of surface defects. Some types of surface, such as those that are highly reflective or transparent, require the implementation of custom-made systems for automated inspections. The purpose of this article is to present a machine vision system, with an easily configurable hardware-software structure, for surface quality inspection of transparent parts. This structure permits that different products and part models may be inspected by the system. Its hardware is composed of image-capturing devices and mechanisms for manipulating and holding the part to be inspected. One such mechanism is the lighting system, which has been specifically developed to allow real-time quality control of this type of surfaces. As regards software design, a component-based approach has been adopted in order to increase reusage of the code and decrease the time required for configuring any type of part and adapting it for inspection. To test the efficiency and robustness of the industrial setup, a series of tests using a transparent industrial part, specifically, a commercial model of headlamp lens, have been assessed. [ABSTRACT FROM AUTHOR]

This paper describes an automated machine vision-based inspection method for fast and accurate detection of broken edges on the machined surface of water pumps which are formed during casting. The paper proposes a collection of steps to inspect images taken in a noisy environment to identify broken edges. By developing three broken edge verification features and using fuzzy C-means clustering, we provide a fuzzy broken edge inspection model that classifies water pumps into three classes: pumps with significant (major) broken edges, pumps with insignificant (minor) broken edges, and pumps with no broken edges. The devised machine vision-based inspection method is efficient in terms of processing time and accuracy to recognize the parts with broken edges and to make decisions consistent with human knowledge. The fuzzy broken edge detection was carried out on a database of 150 gray-level images. The developed method properly identifies about 95 % of the broken edges in the entire database. [ABSTRACT FROM AUTHOR]

In this paper, we describe an ANN with a modified performance function which is used in an automatic inspection system of welding defects in radiographic images. In a first stage, image processing techniques, including noise reduction, contrast enhancement, thresholding and labelling, were implemented to help in the recognition of weld regions and the detection of weld defects. In a second stage, a set of geometrical features which characterise the defect shape and orientation was proposed and extracted between defect candidates. In a third stage, an artificial neural network for weld defect classification was used under a regularisation process with different architectures for the input layer and the hidden layer. Our aim is to analyse this ANN modifying the performance function using a γ parameter in its function, for different neurons in the input and hidden layer in order to obtain a better performance on the classification stage. The automatic system of recognition and classification proposed consists in detecting the four main types of weld defects met in practice plus the non-defect type. The results was compared with the aim to know the parameters that allow the best classification. The correlation coefficients, confusion matrix and the accuracy or the proportion of the total number of predictions that were correct was determined obtaining a value of 80% for the ANN using a modified performance function with a parameter γ=0.6. [ABSTRACT FROM AUTHOR]

This contribution presents a machine vision system capable of revealing, detecting and characterizing defects on non-plane transparent surfaces. Because in this kind of surface, transparent and opaque defects can be found, special lighting conditions are required. Therefore, the cornerstone of this machine vision is the innovative lighting system developed. Thanks to this, the defect segmentation is straightforward and with a very low computational burden, allowing real-time inspection. To aid in the conception of the imaging conditions, the lighting system is completely described and also compared with other commercial lighting systems. In addition, for the defect segmentation, a new adaptive threshold selection algorithm is proposed. Finally, the system performance is assessed by conducting a series of tests using a commercial model of headlamp lens. [ABSTRACT FROM AUTHOR]

The classic image processing method for flaw detection uses one image of the scene, or multiple images without correspondences between them. To improve this scheme, automated inspection using multiple views has been developed in recent years. This strategy's key idea is to consider as real flaws those regions that can be tracked in a sequence of multiple images because they are located in positions dictated by geometric conditions. In contrast, false alarms (or noise) can be successfully eliminated in this manner, since they do not appear in the predicted places in the following images, and thus cannot be tracked. This paper presents a method to inspect aluminum wheels using images taken from different positions using a method called automatic multiple view inspection. Our method can be applied to uncalibrated image sequences, therefore, it is not necessary to determine optical and geometric parameters normally present in the calibrated systems. In addition, to improve the performance, we designed a false alarm reduction method in two and three views called intermediate classifier block (ICB). The ICB method takes advantage of the classifier ensemble methodology by making use of feature analysis in multiple views. Using this method, real flaws can be detected with high precision while most false alarms can be discriminated. [ABSTRACT FROM AUTHOR]

The creation of a non-destructive technique that enables the automatic detection of defects is desirable, and TOFD (Time-Of-Flight Diffraction) technique is gaining rapid prominence due to its high accuracy in detecting, positioning and sizing flaws in steel structures. In this type of imaging, cracks are characterized using sets of hyperbolas, where summit positions correspond to crack tip positions. However, ultrasonic diffracted signals are often too low and difficult to distinguish from noise, and when large structures are inspected, the quantity of data can be extremely large, with the area of interest being very small in comparison to the image size. This paper describes a method that avoids the image formation, replacing it with a sparse matrix (as there is no reason to store and operate on an excessive number of zeros), and automates crack detection by analyzing the curve formed by the sparse matrix elements. The sparse matrix is formed using Split-Spectrum Processing, which enhances the signal-to-noise ratio. The Randomized Hough transform is then applied on the sparse matrix elements to detect the hyperbolas that characterize the crack defects. [ABSTRACT FROM AUTHOR]

A machine vision system that has been developed for the detection of missing fasteners on steel stampings is described. The system has been tested using images generated by a commercial machine vision system installed on an assembly line for the production of automotive cross-car beams. This particular application was considered to be challenging due to variations in the operating conditions, such as the lighting, together with the fact that the appearance of the fasteners could change depending upon their angle relative to the camera. A neuro-fuzzy image classification algorithm was developed and tested against a threshold-based classification algorithm. Results indicate that both algorithms perform well when optimized, but the neuro-fuzzy algorithm was found to degrade in a less abrupt fashion when the input data deviated from the trained data. A graphical user interface (GUI) was designed to enable implementation on the assembly line. The developed system is adaptable to other machine vision-based automated inspection applications. [ABSTRACT FROM AUTHOR]

This paper presents a new edge-detection method which, based on simple arithmetic and logical operations, consists of three procedures: image binarisation, image contraction, and image subtraction. Experiments showed that the proposed algorithm could generate a path one pixel wide with continuous edges, and the proposed algorithm had a better edge-detection accuracy than the 4-connected, 8-connected, and the Sobel techniques. Therefore, the proposed edge-detection algorithm is feasible for use in automatic visual inspection systems. [ABSTRACT FROM AUTHOR]

The industrial application of an automated inspection system that aims to enhance the efficiency and flexibility of a computer integrated manufacturing system (CIMS) is proposed in this paper. A machine-vision-based approach is adopted to utilise its advantages of measurement flexibility, high resolution, and non-destruction. With a closed-loop feedback control architecture, the inspection procedures are fully integrated with the production process to detect and to recover abnormal operations in a real-time mode. Implementation of such a system at a physical shop floor for socket inspection is presented as an example. [ABSTRACT FROM AUTHOR]

The objective of this paper is to document part of a collaborative research program undertaken by the Centre for Computer Integrated Manufacture (CIM Centre) at Swinburne University of Technology in Hawthorn, Victoria, Australia and Australian Defence Industries (ADI) Ltd in the field of 'non-contact inspection'. This research program is one of two collaborative programs between the CIM Centre and ADI Ltd, related to inspecting the quality of components in an automated fashion, without the use of contacting sensors. There are a number of techniques currently being investigated at the CIM Centre, including lasers, vision, acoustic emission and X-ray based methods. This particular paper focuses on the research work undertaken by the first author in the detection of internal surface defects in forged, hollow cylindrical workpieces. The case study presented for consideration and discussion herein is related to the detection of cavity defects in forged pressure vessels. The paper provides a background into the range of alternative non-contact inspection techniques that are available (including lasers, ultrasonics, etc.) and the reasons why some of these failed to provide the functionality that was offered by the vision approach ultimately adopted. There is nothing unique about the application of vision systems in the detection of surface defects. However, in this research program, a number of factors have considerably complicated the application. These include lighting problems and the difficulties encountered in acquiring images within a confined cylindrical space. This paper documents the techniques that have been used to resolve some of these practical image acquisition and processing problems and the relative merits of each approach. The paper also examines some of the algorithmic problems involved in detection of particular surface anomalies in components through a range of techniques and concludes that the one most suited to the surfaces in the pressure vessel case-study is the so-called 'region-growing' approach. [ABSTRACT FROM AUTHOR]

An image registration approach for inspection of printed circuit patterns which has been validated on a prototype system is described. The offline procedure forms, selects, prioritizes, and sorts registration features from CAD-generated reference data. A feature is selected if it satisfies clearance rules that account for the maximum expected discongruence between captured and reference images. The sorting scheme considers the detection complexity of a feature and its distance away from the center of the expected image, since outer features represent potential global distortions better. The runtime registration procedure detects features and finds the parameters that transform pixels into reference data coordinates and vice versa. We represent robust feature-measurement techniques that offer accurate subpixel localization and verify feature authenticity. We describe an edge-detection technique based on a novel way of authenticating zero-crossings and a method that disqualifies edges detected on defects of the part under inspection. [ABSTRACT FROM AUTHOR]