Сегментация дефектов дорожного полотна на основе нейросетевого ансамбля

Railway track defects in Malaysia pose significant risks of train derailments and accidents, underscoring the urgency for early and accurate defect detection and classification. This study presents a novel approach utilizing deep learning models, VGG16 and YOLOv5, for detecting and classifying railway track defects, explicitly focusing on corrugation and squat defects. The research's uniqueness lies in its application of these specific models and the composition of a dataset collected from extensive field measurements and inspections across various railway tracks within the Track Network Maintenance Ampang Line in Malaysia. The results demonstrate that these models achieve high precision in defect classification and detection of defects by more than 80%. The proposed methodology provides the railway industry with a powerful tool to streamline maintenance planning and prioritize defect remediation efficiently. Early defect detection can prevent potential accidents and improve safety and operational efficiency. Future studies can expand on these findings by exploring the extension of the proposed techniques to address other types of rail defects. Incorporating a diverse range of scenarios and operating conditions in the dataset could further enhance the models' performance and generalization. Real-time deployment and integration with existing maintenance systems are crucial for practical adoption. This research has strengths but acknowledges limitations. Additional evaluation metrics and a diverse dataset are essential for model performance. Leveraging deep learning models offers a reliable solution for railway maintenance, enhancing safety and efficiency. Addressing these limitations will drive proactive defect management, ensuring safe and reliable railway networks.

Image processing and analysis techniques have been successfully applied to the evaluation of some types of pavement surface distress, including longitudinal, transverse, and alligator cracking. Many efforts have been made to develop automated evaluation for other types of distress patterns, but, due to the complexities of these distress types, the results were not as positive as for those distresses mentioned above. The main objectives are to develop image segmentation and classification methods to isolate distress features, and to develop the back-propagation neural network model to recognize block cracking and potholes, in addition to alligator, longitudinal, and transverse cracking. It was observed that longitudinal, transverse, alligator, and block cracking were accurately recognized, and the analytical system has a success rate of 93 percent for potholes. Preliminary results indicate that the proposed approach is very positive and has great potential for integration into an automated system for pavem...

Image processing and analysis techniques have been successfully applied to the evaluation of some types of pavement surface distress, including longitudinal, transverse, and alligator cracking. Many efforts have been made to develop automated evaluation for other types of distress patterns, but, due to the complexities of these distress types, the results were not as positive as for those distresses mentioned above. The main objectives are to develop image segmentation and classification methods to isolate distress features, and to develop the back-propagation neural network model to recognize block cracking and potholes, in addition to alligator, longitudinal, and transverse cracking. It was observed that longitudinal, transverse, alligator, and block cracking were accurately recognized, and the analytical system has a success rate of 93 percent for potholes. Preliminary results indicate that the proposed approach is very positive and has great potential for integration into an automated system for pavement distress evaluation.

Software defects and quality assurance are crucial aspects of software development that should be considered during the software development cycle. To ensure high-quality software, it is essential to have a robust quality assurance process in place. System reliability and quality are very key components that must be considered during software development, and this can only be achieved when software undergoes a thorough test process for errors, anomalies, defects, omissions, and bugs. Early software defect prediction and detection play an essential role in ensuring the reliability and quality of software systems, ensuring that software companies discover errors or defects early enough and allocate more resources to defect-prone modules. This study proposes the development of an enhanced classifier model for software defect prediction and detection. The aim is to harness the collective intelligence of selected base classifiers like Support Vector Machine, Logistic regression, Decision Trees, Random Forest, AdaBoost, Gradient Boosting, K-Nearest Neighbor, GaussianNB, and Multi-Layer Perception to improve accuracy, robustness, and generalization in identifying potential defects using a soft voting ensemble technique. The ensemble model leveraged the confidence probability of the soft voting technique and the generalization advantage of cross-validation leading to a more robust and dynamic model. The performance of the model with existing classifiers was evaluated using accuracy, F1 score, Precision, and area under the ROC curve (ROC- AUC) as the evaluation metrics. The results of the experiment revealed that the Proposed Classifier produced an overall Accuracy rate of 93%, and ROC AUC of 98%. The results demonstrate the effectiveness of our enhanced ensemble classifier in software defect detection and prediction. By harnessing the strengths of diverse base classifiers, our approach provides a robust and adaptive solution to the challenges of early detection and mitigating defects in software systems. This research contributes to the advancement of reliable software development practices and lays the foundation for future enhancements in ensemble-based defect detection methodologies.

Demand for wind power has grown, and this has increased wind turbine blade (WTB) inspections and defect repairs. This paper empirically investigates the performance of state-of-the-art deep learning algorithms, namely, YOLOv3, YOLOv4, and Mask R-CNN for detecting and classifying defects by type. The paper proposes new performance evaluation measures suitable for defect detection tasks, and these are: Prediction Box Accuracy, Recognition Rate, and False Label Rate. Experiments were carried out using a dataset, provided by the industrial partner, that contains images from WTB inspections. Three variations of the dataset were constructed using different image augmentation settings. Results of the experiments revealed that on average, across all proposed evaluation measures, Mask R-CNN outperformed all other algorithms when transformation-based augmentations (i.e., rotation and flipping) were applied. In particular, when using the best dataset, the mean Weighted Average (mWA) values (i.e., mWA is the average of the proposed measures) achieved were: Mask R-CNN: 86.74%, YOLOv3: 70.08%, and YOLOv4: 78.28%. The paper also proposes a new defect detection pipeline, called Image Enhanced Mask R-CNN (IE Mask R-CNN), that includes the best combination of image enhancement and augmentation techniques for pre-processing the dataset, and a Mask R-CNN model tuned for the task of WTB defect detection and classification.

Passenger and rail personnel safety is paramount. Rail defects can lead to derailments, collisions, and serious accidents if not detected and addressed in time. A reliable and efficient rail network requires regular and effective infrastructure maintenance. Early defect detection allows for planning and carrying out necessary repairs before problems escalate. Maintenance and repair costs for railway tracks can be significantly reduced through continuous monitoring and prompt intervention when defects are identified. Early defect detection helps extend the lifespan of rails and minimize disruptions to rail traffic, thus improving the overall reliability and availability of the network. Traditional inspection methods, such as visual or ultrasonic checks, have limitations in detecting and characterizing rail defects. Eddy current imaging offers an innovative solution for non-destructive and more comprehensive rail inspection. This technique allows visualizing the rail surface and subsurface in detail, revealing defects that might be difficult to detect using other methods. Imaging provides richer data, enabling in-depth analysis of the size, shape, and location of defects, facilitating accurate assessment of their criticality. Early defect detection through eddy current imaging contributes to informed decision-making in maintenance planning, optimizing interventions and reducing safety risks. This innovative technology is particularly beneficial for high-traffic rail networks, where continuous monitoring and rapid defect detection are essential. In summary, eddy current imaging represents a significant advancement in rail defect detection and characterization, contributing to improved safety, reliability, and efficiency of the rail network.In this paper, a railway inspection system is developed based on the use of multiple sensors for detecting surface defects on the rails. They emphasize the importance of integrating different types of sensors, such as vision sensors, laser sensors, ultrasonic sensors, etc., to achieve a more comprehensive and precise assessment of the rail condition.

Bearings are the basic components of rotating machinery and their integrity is the key to ensuring the operational stability and work reliability of machines. Compared with traditional vibration analysis, acoustic emission (AE) has some unique advantages, such as higher sensitivity to low-speed rotating mechanical defect detection and more potential for early fault detection. However, the AE signals collected from bearing with incipient fault always include heavy noise levels, reducing the capability of early defect detection. Therefore, this paper proposes an optimized Kurtogram method for incipient defect detection of bearings, which combines autocorrelation function, Shannon entropy and Kurtogram to identify early localized defects in AE signals. The major innovations are as follows: (i) The autocorrelation function (ACF) is adopted to process the envelope of all wavelet packet node signals to highlight the periodic pattern in the AE signal, (ii) kurtosis-to-Shannon entropy ratio (KSR) is introduced to improve the capability to detect bearing fault characteristics in low signal-to-noise ratio (SNR) signals. Simulated AE signals and real bearing fault signals were used to evaluate the effectiveness of the proposed method. The results show that the proposed method can detect early defects of bearings and is superior to other Kurtogram-based approaches.

Concrete, known for its high strength, durability, and flexibility, is a core material in construction. However, defects such as voids and honeycombing often occur due to improper pouring or vibration, weakening the concrete's strength and affecting its long-term performance. These defects typically require costly repairs. Therefore, timely identification and repair of such early defects is crucial for improving construction quality. This paper proposes a method for non-destructive detection of honeycomb defects in concrete using infrared thermography (IR) during the hydration stage. By analyzing the temperature differences between defect and non-defect areas based on the temperature distribution generated during hydration, defects can be detected. Furthermore, the study uses the COMSOL finite element model to explore the relationship between defect size, ambient temperature, formwork thickness, and thermal contrast. The results show that IR technology can effectively and reliably detect honeycomb defects, especially during the hydration phase. As a convenient and feasible non-destructive testing method, IR technology has significant potential for application and development in concrete defect detection.

In the last decades, machine vision and Machine Learning (ML) techniques have seen significant improvements in developing new algorithms thanks to the increment of hardware performance. Exploiting machine vision for specific technological applications became an essential opportunity to introduce significant improvements in the manufacturing context. This paper proposes a study to analyze the ML capabilities to perform Automated Optical Inspection (AOI) for quality control in the manufacturing of Printed Circuit Boards (PCBs). The study has been performed by testing Mask R-CNN and YOLOv8 algorithms and an open-source PCB dataset designed to evaluate other ML techniques. The chosen open-source dataset (i.e. PCB defect dataset released by Open Lab on Human–Robot Interaction of Peking University, HRIPCB) individuates appropriate classes of products and related defects for the context of interest, resulting in a suitable dataset for the performance evaluation of tested algorithms. The challenge of this specific application is the recognition of the component boundaries that have submillimetric dimensions and are not clearly identifiable. The comparison between Mask R-CNN and YOLOv8 highlights that the Mask R-CNN performs better in defect detection (i.e., Missing Holes and Shorts). In particular, for the missing hole defects, for example, the mAP50-95 is 0.798 for Mask R-CNN and 0.261 for YOLOv8. Instead, for the short defects, mAP50-95 is 0.519 for Mask R-CNN and 0.399 for YOLOv8. This work has been carried out to gather know-how for further activity related to AOI for quality control in the PCB assembly employed in the aerospace field.

With the rapid development of industrial automation and intelligent manufacturing, defect detection of electronic products has become crucial in the production process. Traditional defect detection methods often face the problems of insufficient accuracy and inefficiency when dealing with complex backgrounds, tiny defects, and multiple defect types. To overcome these problems, this paper proposes Y-MaskNet, a multi-task joint learning framework based on YOLOv5 and Mask R-CNN, which aims to improve the accuracy and efficiency of defect detection and segmentation in electronic products. Y-MaskNet combines the high efficiency of YOLOv5 in target detection with the fine segmentation capability of Mask R-CNN and optimizes the overall performance of the model through a multi-task learning framework. Experimental results show that Y-MaskNet achieves a significant improvement in detection and segmentation tasks, with mAP@[0.5:0.95] reaching 0.72 (up from 0.62 for YOLOv5 and 0.65 for Mask R-CNN) on the PCB Defect Dataset, and IoU improving by 7% compared to existing methods. These improvements are particularly notable in small object detection and fine-grained defect segmentation, making Y-MaskNet an efficient and accurate solution for defect detection in electronic products, offering strong technical support for future industrial intelligent quality control.

[Context] Models play an important role in Software and Systems Engineering processes. Reviews are well-established methods for model quality assurance that support early and efficient defect detection. However, traditional document-based review processes have limitations with respect to the number of experts, resources, and the document size that can be applied. [Objective] In this paper, we introduce a distributed and scalable review process for model quality assurance to (a) improve defect detection effectiveness and (b) to increase review artifact coverage. [Method] We introduce the novel concept of Expected Model Elements (EMEs) as a key concept for defect detection. EMEs can be used to drive the review process. We adapt a best-practice review process to distinguish (a) between the identification of EMEs in the reference document and (b) the use of EMEs to detect defects in the model. We design and evaluate the adapted review process with a crowdsourcing tool in a feasibility study. [Results] The study results show the feasibility of the adapted review process. Further, the study showed that inspectors using the adapted review process achieved results for defect detection effectiveness, which are comparable to the performance of inspectors using a traditional inspection process, and better defect detection efficiency. Moreover, from a practical perspective the adapted review process can be used to complement inspection efforts conducted using the traditional inspection process, enhancing the overall defect detection effectiveness. [Conclusions] Although the study shows promising results of the novel process, future investigations should consider larger and more diverse review artifacts and the effect of using limited and different scopes of artifact coverage for individual inspectors.

The prompt and precise detection of road damage is vital for effective infrastructure management, forming the foundation for intelligent transportation systems and cost-effective pavement maintenance. While current convolutional neural network (CNN)-based methodologies have made progress, they are fundamentally limited by treating damages as independent, isolated entities, thereby ignoring the intrinsic spatial symmetry and topological organization inherent in complex damage patterns like alligator cracking. This conceptual asymmetry in modeling leads to two major deficiencies: “context blindness,” which overlooks essential structural interrelations, and “temporal inconsistency” in video analysis, resulting in unstable, flickering predictions. To address this, we propose a Spatio-Temporal Graph Mamba You-Only-Look-Once (STG-Mamba-YOLO) network, a novel architecture that introduces a symmetry-informed, hierarchical reasoning process. Our approach explicitly models and integrates contextual dependencies across three levels to restore a holistic and consistent structural representation. First, at the pixel level, a Mamba state-space model within the YOLO backbone enhances the modeling of long-range spatial dependencies, capturing the elongated symmetry of linear cracks. Second, at the object level, an intra-frame damage Graph Network enables explicit reasoning over the topological symmetry among damage candidates, effectively reducing false positives by leveraging their relational structure. Third, at the sequence level, a Temporal Graph Mamba module tracks the evolution of this damage graph, enforcing temporal symmetry across frames to ensure stable, non-flickering results in video streams. Comprehensive evaluations on multiple public benchmarks demonstrate that our method outperforms existing state-of-the-art approaches. STG-Mamba-YOLO shows significant advantages in identifying intricate damage topologies while ensuring robust temporal stability, thereby validating the effectiveness of our symmetry-guided, multi-level contextual fusion paradigm for structural health monitoring.

In recent decades, millions of people are killed by natural disasters such as wildfire, landslide, tsunami, and volcanic eruption. The efficiency of post-disaster emergency responses and humanitarian assistance has become crucial in minimizing the expected casualties. This paper focuses on the task of building damage level evaluation, which is a key step for maximizing the deployment efficiency of post-event rescue activities. In this paper, we implement a Mask R-CNN based building damage evaluation model with a practical two-stage training strategy. The motivation of Stage-l is to train a ResNet 101 backbone in Mask R-CNN as a Building Feature Extractor. In Stage-2, we further build on top the model trained in Stage-l a deep learning architecture that performs more sophisticated tasks and is able to classify buildings with different damage levels from satellite images. In particular, in order to take advantage of pre-disaster satellite images, we extract the ResNet 101 backbone from the Mask R-CNN trained on pre-disaster images in Stage-l and utilize it to build a Siamese based semantic segmentation model for classifying the building damage level at the pixel level. The pre- and post-disaster satellite images are simultaneously fed into the proposed Siamese based model during the training and inference process. The output of these two models own the same size as input satellite images. Buildings with different damage levels, i.e., ‘no damage’, ‘minor damage’, ‘major damage’, and ‘destroyed’, are represented as segments of different damage classes in the output. Comparative experiments are conducted on the xBD satellite imagery dataset and compared with multiple state-of-the-art methods. The experimental results indicate that the proposed Siamese based method is capable to improve the damage evaluation accuracy by 16 times and 80%, compared with a baseline model implemented by xBD team and the Mask-RCNN framework, respectively.

The quality of the software design often has a major impact on the quality of the final product and the effort for development and evolution. A number of quality assurance (QA) approaches for inspection of early-life-cycle documents have been empirically evaluated. An implicit assumption of these studies was: an investment into early defect detection and removal saves higher rework cost. The concept of pair programming combines software construction with implicit QA in the development team. For planning QA activities, an important research question is how effective inspectors can be expected to be at detecting defects in software (design and code) documents compared to programmers who find defects as by-product of their usual construction activities. In this paper we present an initial empirical study that compares the defect detection effectiveness of a best-practice inspection technique with defect detection as by-product of constructive software evolution tasks during pair programming. Surprisingly, in the study context pair programmers were more effective to find defects in design documents than inspectors. However, when building a larger team for defect detection, a mix of inspection and pair programming can be expected to work better than any single technique.

Diagnosis and fault detection in mechanical systems during their time-varying non stationary operation is one of the most challenging tasks. The paper presents a method for the early detection of gearbox defects based on the empirical mode decomposition (EMD) algorithm and a proposed modified Hilbert transform. The EMD technique decomposes the measured signal into oscillatory functions called Intrinsic Mode Functions (IMF). A numerical model of damaged gears is used for generating a modulated vibratory signal with repetitive shocks. The application of time descriptors “Talaf” and “Thikat” to different IMF decomposition levels of the modified Hilbert envelope gives good results for early detection of defects in comparison with the IMF of the original time signal and its traditional Hilbert envelope or with the wavelet decomposition.