Class Of Defects Research Articles

A Deep Learning Approach to Classify and Detect Defects in the Components Manufactured by Laser Directed Energy Deposition Process

This paper presents a deep learning approach to identify and classify various defects in the laser-directed energy manufactured components. It mainly focuses on the Convolutional Neural Network (CNN) architectures, such as VGG16, AlexNet, GoogLeNet and ResNet to perform the automated classification of defects. The main objectives of this research are to manufacture components using the laser-directed energy deposition process, prepare a dataset of horizontal wall structure, vertical wall structure and cuboid structure with three defective classes such as voids, flash formation, and rough textures, and one non-defective class, use this dataset with a deep learning algorithm to classify the defect and use the efficient algorithm to detect defects. The next objective is to compare the performance parameters of VGG16, AlexNet, GoogLeNet and ResNet used for classifying defects. It has been observed that the best results were obtained when the VGG16 architecture was applied to an augmented dataset. With augmentation, the VGG16 architecture gave a test accuracy of 94.7% and a precision of 80.0%. The recall value is 89.3% and an F1-Score is 89.5%. The VGG16 architecture with augmentation is highly reliable for automating the defect detection process and classifying defects in the laser additive manufactured components.

Laser welding defects detection in lithium-ion battery poles

Strong demand for electric vehicles and energy storage applications has led to a rapid expansion of the battery sector. Laser welding is widely used in lithium-ion batteries and manufacturing companies due to its high energy density and capability to join different materials. Welding quality plays a vital role in the durability and effectiveness of welding structures. Therefore, it is essential to monitor welding defects to ensure welds quality. Manual inspection, analysis and evaluation of welding defect images is difficult due to the non-uniformity in their shape, position, and size. Hence the use of deep learning techniques to identify welding defects is more accurate and reliable due to the adequate training data samples, which helps to identify welding defects with greater accuracy. In this study, we present a novel collection of 3,736 laser welding images which are labeled with eight classes. This dataset contains both normal and defective classes collected from a Dade Laser Chinese production line. Moreover, we introduce a modified loss function that integrates cross entropy and complement objective training. The EfficientNetB6 model is trained with the newly introduced loss function that performed efficiently and achieved a promising accuracy during the classification phase. The experimental results demonstrate the significance of the proposed loss function and the sufficiency of EfficientNetB6 model on the laser welding dataset. Compared to other models, EfficientNetB6 performs better.

An Online Quality Detection Method With Ensemble Learning on Imbalance Data for Wave Soldering

Abstract Online detection of wave soldering is an important method of inspecting defective products in the workshop. Accurate quality detection can reduce production costs and provide support for quality warnings in the wave soldering process. However, there are still problems in improving the detection accuracy for the defect class. Although class imbalance in data can be addressed by data-level methods such as over-sampling and under-sampling, these methods destroy the integrity of the original data set and may cause information loss and over-fitting problems. In order to solve the above problems, this article focuses on how to design a new loss function that fuses class weights from focal loss (FS) and sample weights from AdaBoost to improve attention to the minority samples without changing data distribution. In this way, an FS-AdaBoost-RegNet model based on transfer learning is constructed to enhance the detection accuracy in industrial environment. Finally, the images of the wave soldering from an electronic assembly workshop are taken to validate the performance of the proposed method. The experiment on 941 testing samples of the imbalance datasets showed that the FS-AdaBoost-RegNet model with new loss function reached the overall accuracy of 98.39%, and the overall recall of 96.19%. The results proved that the proposed method promotes the ability to identify defect class compared with other methods.

A Model Classifying Four Classes of Defects in Reinforced Concrete Bridge Elements Using Convolutional Neural Networks

Recently, the bridge infrastructure in Ukraine has faced the problem of having a significant number of damaged bridges. It is obvious that the repair and restoration of bridges should be preceded by a procedure consisting of visual inspection and evaluation of the technical condition. The problem of fast and high-quality collection, processing and storing large datasets is gaining more and more relevance. An effective way to solve this problem is to use various machine learning methods in bridge infrastructure management. The purpose of this study was to create a model based on convolutional neural networks (CNNs) for classifying images of concrete bridge elements into four classes: “defect free”, “crack”, “spalling” and “popout”. The eight CNN models were created and used to conduct its training, validation and testing. In general, it can be stated that all CNN models showed high performance. The analysis of loss function (categorical cross-entropy) and quality measure (accuracy) showed that the model on the MobileNet architecture has optimal values (loss, 0.0264, and accuracy, 94.61%). This model can be used further without retraining, and it can classify images on datasets that it has not yet “seen”. Practical use of such a model allows for the identification of three damage types.

Studying the effectiveness of deep active learning in software defect prediction

Accurate prediction of defective software modules is of great importance for prioritizing quality assurance efforts, reasonably allocating testing resources, reducing costs and improving software quality. Several studies have used machine learning to predict software defects. However, complex structures and imbalanced class distributions in software defect data make learning an effective defect prediction model challenging. In this article, two deep learning-based defect prediction models using static code metrics are proposed. In order to enhance the learning process and improve the performance of the proposed models, pool-based active learning is employed. In this regard, the possibility of using active learning to mitigate the need for a large amount of labeled data in the process of building deep learning models is investigated. To deal with imbalanced distribution of software modules between defective and non-defective classes, Near-Miss under-sampling and KNN, with different number of neighbors, are used. The reason for choosing them is their good performance in binary classification problems. Experiments are performed on two well-known, publicly available datasets, GitHub Bug Dataset and public Unified Bug Dataset for java projects. The evaluation results reveal the effectiveness of our proposed models in comparison to the traditional machine learning algorithms. In the conducted investigations on the Unified Bug Dataset, at the file level, the value of F-measure and AUC criteria have improved by 13 and 11 percent, respectively and at the class level, the values have improved by 14 and 11 percent, respectively.

Process Defects Knowledge Modeling in Laser Powder Bed Fusion Additive Manufacturing: An Ontological Framework

Comprehensive knowledge of the laser powder bed fusion (LPBF) process defects, their causal factors, and relationships can enable proactive prevention and/or mitigation of those defects to ensure the production of high-quality products. However, this knowledge is scattered in a plethora of research articles, and there is a need for a formal and structured knowledge base to document the LPBF process defects knowledge and model the complex network of relationships among those defects and their causal factors. In response, this paper proposes an ontological framework to systematically structure and represent the knowledge of LPBF defects in a sustainable, reusable, and extensible way. In doing so, we first conducted a detailed literature review and analysis of current LPBF defect surveys to systematically develop a consistent and comprehensive classification of defects and potential causal factors. Then, to effectively represent the gathered knowledge, the ontological framework was designed to: (1) organize and formalize knowledge on LPBF defects and the causal factors, (2) model the complex network of causal links and cascading effects among the defects and causal factors, (3) enable easy querying of the stored knowledge, and (4) include ontological entities that are suitable for extension and reuse. A prototype LPBF defects ontology was developed in Web Ontology Language (OWL)/Resource Description Framework (RDF) formalism using the Protégé tool to effectively realize those design requirements. The developed ontology covers thirty-one unique defects and knowledge of their causal factors, including defect-to-defect causal relationships and hierarchical categorization under four major categories of high-level defect classes. Similarly, forty-five unique causal factors were categorized under twelve major categories. The proposed ontological framework and knowledge model offer a pathway to (1) provide a comprehensive knowledge base on LPBF defects for in-depth tutoring and training of novice researchers and practitioners; (2) help investigators to identify root causes of detected defects for informed corrective action; (3) guide process planning tasks from a defect control perspective; and (4) support future application and reuse of the knowledge. This study also provides several examples to illustrate the modeling of cascading effects in causal relationships, discovering knowledge using an ontology reasoner, visualizing the complex network of causal links using OntoGraph, and retrieving stored knowledge by answering competency questions using SPARQL queries. In the future, the reasoning capabilities of our proposed ontology can also be leveraged to develop expert systems for optimizing the AM workflow and quantitatively predict and diagnose LPBF defects.© 2023 Society of Manufacturing Engineers (SME). Published by Elsevier Ltd. All rights reserved.This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)Peer-review under responsibility of the Scientific Committee of the NAMRI/SME.

Detecting Surface Defects of Achacha Fruit (Garcinia humilis) with Hyperspectral Images

Hyperspectral imaging data within the wavelength range of 400–1000 nm were used to classify the common skin conditions (i.e., normal, scar, decay, and insect bite) of achacha fruits. The band ratio (BR) and spectral angle mapper (SAM) algorithms were used in a binary classification. Furthermore, SAM, support vector machine (SVM), and artificial neural network (ANN) models were used in a multiclass classification. The performances of the binary and multiclass classification models were assessed. For the binary-classification approach, the three defective classes were merged into one, and the accuracies of the BR (990 nm/600 nm) and SAM were 78.70% and 75.02%, respectively. Furthermore, the SAM, SVM, and ANN accuracies in the four class problems were 58.36%, 83.59%, and 99.88%, respectively. A principal component analysis (PCA) was used for the data reduction. Nine characteristic wavelengths were extracted from the weighting-coefficient curves of the first four principal components. Using only the nine selected bands, the accuracies of the SAM, SVM, and ANN models were 51.49%, 80.76%, and 96.85%, respectively. Compared with the models using full bands, the classification accuracies of the models using only nine characteristic bands decreased slightly; however, the gain in classification speed and the potential data-acquisition speed can expedite the classification of achacha fruits.

Finding defects in glasses through machine learning

Structural defects control the kinetic, thermodynamic and mechanical properties of glasses. For instance, rare quantum tunneling two-level systems (TLS) govern the physics of glasses at very low temperature. Due to their extremely low density, it is very hard to directly identify them in computer simulations. We introduce a machine learning approach to efficiently explore the potential energy landscape of glass models and identify desired classes of defects. We focus in particular on TLS and we design an algorithm that is able to rapidly predict the quantum splitting between any two amorphous configurations produced by classical simulations. This in turn allows us to shift the computational effort towards the collection and identification of a larger number of TLS, rather than the useless characterization of non-tunneling defects which are much more abundant. Finally, we interpret our machine learning model to understand how TLS are identified and characterized, thus giving direct physical insight into their microscopic nature.

Transpyloric Tube Placement Shortens Time to Full Feeding in Left Congenital Diaphragmatic Hernia

BackgroundNutritional complications have an impact in both short- and long-term morbidity of patients with congenital diaphragmatic hernia (CDH). We aimed to compare time to full enteral tube feeding depending on route -gastric (GT) or transpyloric (TPT)- in newborns with left CDH (L-CDH). MethodsRetrospective cohort study of L-CDH patients admitted to a referral tertiary care NICU between January 2007 and December 2014. Lethal chromosomal abnormalities and death before initiation of enteral nutrition were exclusion criteria. Results37 patients were fed through GT, 46 by TPT. TPT children took 11.0 (6.8) days to reach full enteral tube feeding and spent 16.6 (8.1) days on parenteral nutrition vs 16.8 (14.7) days (p = 0.041) and 22.7 (13.5) days (p = 0.020) of GT patients. TPT children had 3.9 (2.4) days of fasting due to GI issues and 20% had episodes of decreased rates of enteral nutrition for extra-GI complications vs 11.4 (11.1) days (p = 0.028) and 49% (p = 0.006). According to the best fitting model (R2 0.383, p < 0.001), the TPT-group achieved full enteral feeding 8.4 days earlier than the GT-group (95% CI -14.76 to - 2.02 days), after adjustment by severity of illness during the first days, o/e LHR_liver and class of diaphragmatic defect. There were no differences in growth outcomes and length of stay between survivors of GT and TPT groups. ConclusionTPT shortens time to full enteral nutrition, especially in the most severe L-CDH patients. We propose that placement of a TPT at the end of the surgical repair procedure should be considered, especially in higher-risk patients. Level of EvidenceTreatment study, Level III. Retrospective comparative, case–control study.

Correlation of Extended Defects with Electrical Yield of SiC MOSFET Devices

The quality of the silicon carbide (SiC) epitaxial layer, i.e., layer homogeneities and extended defect densities, is of highest importance for high power 4H-SiC trench metal-oxide-semiconductor field effect transistors (Trench-MOSFET) devices. Especially, yield for devices with a large chip area is severely impacted by extended defects. Previously, devices had to be fully manufactured to effectively gauge the impact of a reduction in extended defect densities in the epitaxial layers on device yield. The production of devices such as Trench-MOSFETs is an extensive procedure. Therefore, a correlation between extended defects in the epitaxial layer and electrical device failure would allow to reliably estimate the impact of process changes during epitaxial layer deposition on electrical device yield.For this reason, n-type epitaxial layers were grown on around 1,000 commercially available 150 mm 4H-SiC Si-face substrates, which received a chemical wet cleaning prior to the epitaxy deposition. Substrates with lowest micro-pipe density from two different suppliers were used. The wafers were characterized with the corresponding device layout for defects utilizing surface microscopy as well as ultraviolet photoluminescence techniques. Subsequently, these wafers were used to produce more than 500,000 Trench-MOSFET devices. All devices have been tested on wafer level for their initial electrical integrity.With these methods a precise correlation between extended defects in the epitaxial layer and electrical failures on wafer level could be found. The influence of different substrates on the defect-based yield prediction regarding the electrical yield on wafer level is discussed. Additionally, a calculated kill-ratio is presented and the severity of defect classes on initial device failure, e.g., stacking faults, and their key failures modes are discussed.

Electrochemical Doping of Halide Perovskites by Noble Metal Interstitial Cations.

Metal halide perovskites are an attractive class of semiconductors, but it has proven difficult to control their electronic doping by conventional strategies due to screening and compensation by mobile ions or ionic defects. Noble-metal interstitials represent an under-studied class of extrinsic defects that plausibly influence many perovskite-based devices. In this work, doping of metal halide perovskites is studied by electrochemically formed Au+ interstitial ions, combining experimental data on devices with a computational analysis of Au+ interstitial defects based on density functional theory (DFT). Analysis suggests that Au+ cations can be easily formed and migrate through the perovskite bulk via the same sites as iodine interstitials (Ii + ). However, whereas Ii + compensates n-type doping by electron capture, the noble-metal interstitials act as quasi-stable n-dopants. Experimentally, voltage-dependent, dynamic doping by current density-time (J-t), electrochemical impedance, and photoluminescence measurements are characterized. These results provide deeper insight into the potential beneficial and detrimental impacts of metal electrode reactions on long-term performance of perovskite photovoltaic and light-emitting diodes, as well as offer an alternative doping explanation for the valence switching mechanism of halide-perovskite-based neuromorphic and memristive devices.

Hierarchical Neural Network and Simulation Based Structural Defect Identification and Classification

A vibration data-driven structural defect identification and classification technique is developed using frequency response under random excitation and a hierarchical neural network. A system of artificial neural networks (ANNs) is trained using finite element simulation-based synthetic data to reduce the need for many sensor measurements required otherwise. Principal component analysis (PCA) is employed to compress the high dimensionality of the vibration response data and eliminate the noise effect in the training and testing. Frequency responses data dimension for the structure with defects such a crack from stress concentration, rivet hole expansion, and attached foreign object mass such as ice accumulation in aircraft wing or fuselage are reduced using PCA and fed to a classifier network. The probabilistic decision output from the classifier network and the compressed data are then fed to the next levels of estimator networks, where each network is dedicated to the individual type of defect for the estimation of the defect parameters corresponding to that class of defect. The methodology is applied to a stiffened panel structure. The cracks and rivet hole expansions are introduced in the rivet line of the stiffener, and the foreign object mass is attached to the panel surface. The results show that it is possible to classify the defects and further estimate the defect parameters with good accuracy and reliability. It was observed that the damage classification network had an accuracy of roughly 95%. The damage localization network for crack as well as rivet expansion had average absolute error of around 2. The damage severity network was also able to perform well with a mean absolute error of about 0.34 for crack length detection and 0.22 for expanded rivet damage. However, the damage localization and severity prediction networks were quite challenging to train in the presence of multiple damages and need further development in the network architecture.

Few-shot incremental learning in the context of solar cell quality inspection

Deep Neural Networks have shown high defect detection rates in industrial setups, surpassing other more traditional manual feature engineering-based proposals. This has been mainly achieved through supervised training, where a great number of annotated samples are required to learn good classification models. However, obtaining such a large amount of data is sometimes challenging in industrial scenarios, as defective samples do not occur regularly, and certain types of defects only appear occasionally. In this work, we explore the technique of weight imprinting in the context of solar cell quality inspection. This technique allows to incorporate new classes into the classification network using just a few samples. We tested the technique by first training a base network for the segmentation of three base defect classes and then sequentially incorporating two additional defect classes. This resulted in a network capable of segmenting five different defect classes. We also experimented with the network architecture, resulting in more precise segmentation and defect detection results. The experiments’ results have shown that this technique allows the network to extend its capabilities regarding the detection of new defect classes using just a few samples, which can be interesting for industrial practitioners.

Codimension-2 defects and higher symmetries in (3+1)D topological phases

(3+1)D topological phases of matter can host a broad class of non-trivial topological defects of codimension-1, 2, and 3, of which the well-known point charges and flux loops are special cases. The complete algebraic structure of these defects defines a higher category, and can be viewed as an emergent higher symmetry. This plays a crucial role both in the classification of phases of matter and the possible fault-tolerant logical operations in topological quantum error-correcting codes. In this paper, we study several examples of such higher codimension defects from distinct perspectives. We mainly study a class of invertible codimension-2 topological defects, which we refer to as twist strings. We provide a number of general constructions for twist strings, in terms of gauging lower dimensional invertible phases, layer constructions, and condensation defects. We study some special examples in the context of \mathbb{Z}_2ℤ2 gauge theory with fermionic charges, in \mathbb{Z}_2 \times \mathbb{Z}_2ℤ2×ℤ2 gauge theory with bosonic charges, and also in non-Abelian discrete gauge theories based on dihedral (D_nDn) and alternating (A_6A6) groups. The intersection between twist strings and Abelian flux loops sources Abelian point charges, which defines an H^4H4 cohomology class that characterizes part of an underlying 3-group symmetry of the topological order. The equations involving background gauge fields for the 3-group symmetry have been explicitly written down for various cases. We also study examples of twist strings interacting with non-Abelian flux loops (defining part of a non-invertible higher symmetry), examples of non-invertible codimension-2 defects, and examples of the interplay of codimension-2 defects with codimension-1 defects. We also find an example of geometric, not fully topological, twist strings in (3+1)D A_6A6 gauge theory.

Semi-supervised rotation-invariant representation learning for wafer map pattern analysis

Recently, data-driven approaches have been widely employed to analyze the defect patterns in wafer maps, which are crucial for identifying the root causes of failures in the semiconductor fabrication process. Representation learning embeds wafer maps into compact vector representations of useful features, based on which various downstream tasks can be performed to efficiently analyze the patterns on a large scale. If wafer maps are annotated with their defect class labels, the learned representations of wafer maps will be more informative and discriminative in defect patterns. However, the manual labeling of all wafer maps by domain experts is difficult due to practical constraints. In this study, we present a semi-supervised representation learning method that fully utilizes the information from both unlabeled and labeled wafer maps to learn better representations of wafer maps with a lower labeling cost. Given a partially labeled dataset, rotation-invariant representations of wafer maps are learned using the following three objectives. First, each unlabeled wafer map is close to any wafer map of a certain class and far from those of other classes. Second, each pair of labeled wafer maps are close to each other if they belong to the same class and are far from each other otherwise. Third, the different rotations of each wafer map are close to each other for both the unlabeled and labeled wafer maps. The effectiveness of the proposed method is demonstrated for various downstream tasks related to wafer map pattern analysis: visualization, clustering, retrieval, and classifier training.

Faulty voice diagnosis of automotive gearbox based on acoustic feature extraction and classification technique

This research work objectives to establish a systematic approach to automobile gearbox defect diagnostics by acquiring the acoustic feature from their faulty sound information using efficient machine learning techniques. Machine learning method is important for automated fault diagnosis because it uses acquired data to identify machine conditions and defects from their sound. This motivates studying machine learning for gearbox defect diagnosis using acoustic analysis, and their sound inputs. Unexpected gear and bearing failures result in considerable system damage, significant financial loss, and significant safety concerns, in addition to significant system damage. The constant operation and fluctuating cyclic loading conditions can result in catastrophic failure of the gearbox. Utilising condition monitoring, predictive maintenance of a gearbox has been practised over time. Numerous methods, including acoustic analysis from their normal sound and some faulty sound in gear box and others, can be used to conduct non-destructive testing. Following the creation of an experimental setup for several classes of gearbox defects, acoustic features are extracted from the gear in the form of statistical coefficients in term of mean and correlation coefficients. The acoustic features are calculated using Mel-frequency cepstrum coefficients (MFCC) and their derivatives ∆MFCC and ∆∆ MFCC feature extraction technique. The best statistical features are chosen using support vector machine (SVM), decision tree (DT), linear decrement analysis (LDA), Naive Bayes (NB) and Logarithm regression (LR) classifiers. All of the mentioned classifiers here SVM and NB are given the best classification result. From the features based classifiers techniques, the defective component was found.

AI-Powered Multi-Class Defect Segmentation in Industrial CT Data

Industrial computed tomography is a key technology for non-destructive testing for many applications. The growing demand of automated testing requires an efficient automated data analysis. In this work we propose a solution using a neural network for voxel-accurate segmentation of 4 relevant defect classes in industrial casting parts and additively manufactured parts. Defect classes are pores, cracks, inclusions and pore nests. The network was trained and evaluated with a dataset of 20,000 samples. We used a modified U-Net architecture originally developed for medical image segmentation which we optimized for our industrial CT-Data and the multi-class segmentation task. In this work we show that we were able to achieve promising results of ~82 % MeanIoU score and individual defect class MeanIoU values of up to 95 % segmentation performance and qualitatively a high classification performance.

Topological defects in silicene

Using the molecular dynamic simulations a new class of topological defects in silicene is investigated. Si atoms in silicene belong to two triangular sublattices shifted one with respect to the other by h along the direction normal to the sheet. Silicene can have two energetically equivalent structures when the first or second sublattice is above the other. Topological defects appear at the junctions of the domains of these two structures. Domain walls can be rectilinear or curvilinear. Such defects cannot disappear from the structure except as a result of the annihilation of defects with positive and negative topological charges. Structure and energy of the topological defects are calculated. The effect of temperature is shown. Similar defects inevitably exist in other group IVA elemental 2D materials with buckled structure, such as germanene, stanene, and plumbene. As a result of the work, the problem of experimental detection of such defects and the question of how they affect the physicochemical and mechanical properties of materials, in particular, toxicity or sensory properties, is posed.

An application of Generative Adversarial Networks to improve automatic inspection in automotive manufacturing

In manufacturing systems, the quality of inspection is a critical issue. This can be conducted by humans or by employing Computer Vision Systems (CV S), which are trained upon representative datasets of images to detect classes of defects that may occur. The construction of such datasets strongly limits the use of CV S methods, as the variety of defects has combinatorial nature. Alternatively, instead of recognizing defects, a system can be trained to detect non-defective standards, becoming appropriate for some application profiles. In automotive manufacturing, for example, parts are assembled within a reduced set of correct combinations, while the amount of possible incorrect assembling is enormous. This paper integrates a Generative Adversarial Network (GAN) within the CV S framework used by Renault/Brazil to improve the detection of defective production in its automotive assembly line. By sparing the construction of expensive defect image datasets, our solution has proved to be cost-effective and more efficient in comparison with the current CV S solution to detect defects, besides generalizing better to inspect different components without any modification in the method.

MSRConvNet: Classification of railway track defects using multi-scale residual convolutional neural network

The development of an automated rail line defect classification system is of great benefit, as railway tracks must be periodically monitored and inspected to guarantee the safety of rail transportation. In this paper, an effective multi-scale residual convolutional network (MSRConvNet) model is proposed to classify the different types of railway track defects. The skip connections with residual learning blocks are used to increase the effectiveness of the network. The multi-scale convolutions are connected with parallel and two skip connections in the structure to distribute detailed feature maps with each other. Therefore, different scale feature maps can be extracted. The data augmentation method is performed to ensure a balanced class distribution and to eliminate the negative effect of the imbalanced dataset. The proposed model is compared with both benchmark deep learning models and the different variations of the designed network. The results verify that the proposed model can reach superior classification fulfillment, and the MSRConvNet provides an overall accuracy of 99.83%, precision of 99.83%, sensitivity of 99.83%, specificity of 99.94%, F1-score of 99.83%, and Matthew’s correlation coefficient of 99.78% for four defect classes.