Computers in Industry最新文献

In industrial sectors such as shipping, chemical processing, and energy production, centrifugal pumps often experience failures due to harsh operational environments, making it challenging to accurately identify fault types. Traditional fault diagnosis methods, which heavily rely on existing fault datasets, suffer from limited generalization capabilities, especially when substantial labeled and specific fault sample data are lacking. This paper proposes a novel fault diagnosis approach for centrifugal pumps, utilizing a digital twin (DT) framework powered by a graph transfer learning model to address this issue. Firstly, a high-fidelity DT model is constructed to simulate the flow-induced vibration response of the impeller under different health states to enrich the type and scale of the dataset. Secondly, a graph convolutional neural networks (GCN) model is constructed to learn the knowledge of simulation data, and the Wasserstein distance between simulation data and measured data is optimized for adversarial domain adaptation, thereby achieving efficient cross-domain fault diagnosis. Experimental results demonstrate that the proposed algorithm delivers effective fault diagnosis with minimal prior knowledge and outperforms comparable models. Furthermore, the DT system developed using the proposed model enhances the operational reliability of centrifugal pumps, reduces maintenance costs, and presents an innovative application of DT technology in industrial fault diagnosis.

The accuracy and quality of part formation are crucial considerations. However, the laser directed energy deposition (L-DED) process often leads to irregular changes in deposition contours and mechanical properties across parts due to complex flow fields and temperature variations. Hence, to ensure the forming accuracy and quality, it is necessary to achieve precise monitoring and appropriate parameter adjustments during the processing. In this study, a machine vision method for real-time monitoring is proposed, which combines target tracking and image processing techniques to achieve accurate recognition of deposition contours under noisy conditions. Through comparative verification, the measurement accuracy reaches as high as 98.98 %. Leveraging the monitoring information, a bidirectional prediction neural network is proposed to accomplish layer-by-layer forward prediction of layer height. Meanwhile, inverse prediction is employed to determine the processing parameters required for achieving the desired layer height, facilitating the optimization of the deposition contours. It was found that as the processing parameters were adjusted layer-by-layer to achieve consistent deposition contours, there was also a tendency towards consistent changes in microstructure and mechanical properties. The standard deviations of primary dendrite arm spacing (PDAS) and ultimate tensile strength (UTS) at different positions decrease by over 52.2 % and 61.4 %, respectively. This study reveals the consistent patterns of variation in deposition contours and mechanical properties under data-driven variable parameter processing, laying an important foundation for future exploration of the complex process-structure-performance (PSP) relationship in L-DED.

Improvements in the performance and graphical quality of Head-Mounted Displays (HMDs) have led to their increasing use in Virtual Reality (VR) for product presentation and virtual prototype (VP) evaluations. Various locomotion techniques in VR make it possible to move through a virtual scenario and approach the VP for evaluation purposes. The integration of eye-tracking devices into recent HMDs allows the trajectory and gaze behavior of observers to be reported during the evaluation, often more objectively than self-report questionnaires. However, very few studies have used physiological measures for the evaluation of products embedded in VR environments. Therefore, this paper offers a study in which 95 people evaluated three VPs of street furniture presented in their environment of use using Meta Quest Pro headset and explored through teleport and natural walking. The influence of the locomotion techniques on the ratings recorded using a semantic differential, sense of presence, cybersickness, and the role of eye-tracking in understanding gaze behavior while evaluating products' Areas of Interest (AOIs), are investigated. This study found no evidence that the way of approaching the product influences the evaluation of some of its features, overall product evaluation, confidence in responses, sense of presence, or cybersickness differently. On the other hand, this work evidences that the locomotion technique had an impact on how the user approached the products, which could significantly influence the viewing time of some AOIs. The study revealed that the most observed AOIs coincided with those parts closely related to important features, generally located at the top of the products, so paying special attention to these parts when designing and evaluating similar VPs is recommended.

A metal tube system is known as the industrial blood vessel, among which the bent section is the most vulnerable part. The cross-sectional defects (CSDs) of the bent tube cause the flow fluctuation of the fluid inside the tube. The existing defect characterization methods are roughly presented by describing CSDs in some specific cross-sections, which results in the lack of the tube full-bent section (FBS) characteristic information. To comprehensively describe and predict the tube FBS characteristics, an advanced physics-embedded CSDs prediction framework is proposed. This framework includes an FBS-neutral layer displacement angle (NLDA) prediction module and an FBS-CSDs prediction module, which uses the method that integrates the analytical model and BiLSTM-based deep learning (DL) models to predict the CSDs in the FBS of the tube. A novel analytical model of CSDs that considers both three-directional stresses and strains during tube bending is embedded in the FBS-CSDs prediction module. The analytical model provides the initial predicted values of CSDs through the NLDA sequence obtained from the FBS-NLDA module. The inaccurate CSDs are then treated as physical information to be fed into DL models for further correction and prediction. The prediction performance of this framework is validated through numerical simulations and experiments. The results prove that the framework can accurately predict the CSDs in the tube FBS. The integration of DL models with the analytical model not only overcomes the limitations of the analytical model, but also improves the prediction accuracy and convergence speed of DL models.

Critical infrastructures (CIs) are highly susceptible to cyber threats due to their crucial role in the nation and society. Intrusion Detection Systems (IDS) are deployed at the process level to enhance CI security. These process-level IDSs are broadly categorized into univariate and multivariate systems. Our research underscores that both types of systems encounter limitations, especially in handling correlations among process variables (PVs). Univariate IDSs neglect correlations by assessing PVs in isolation, while multivariate IDSs capture these but are vulnerable to evasion attacks. In response, we introduce ProIDS- a novel segmentation and segregation-based process-level IDS. ProIDS leverages the inherent correlations among PVs while segregating them into distinct units to enhance security against evolving threats. This strategic approach ensures the capture of correlations and mitigates the risk of evasion attacks, enhancing the system’s ability to detect abnormal activities. Additionally, ProIDS offers non-parametric modeling for heightened performance, minimal computational overhead, and noise reduction properties. Our comprehensive experiments demonstrate ProIDS’s superiority over baseline methods, delivering precise detection of various attacks while maintaining operational efficiency.

The radiographic inspection plays a crucial role in ensuring the casting quality for improving the service life under harsh environments. However, due to the low-contrast between the defects and the image background, the random spatial position distribution, random shapes and aspect ratios of the defects, the development of an accurate defect automatic detection system is still challenging. To address these issues, this paper proposes a novel framework for low-contrast and random multi-scale casting defect detection, which is referred to as adaptive global dynamic detection transformer (AGD-DETR). A novel defect-aware data augmentation method is first proposed to adaptively highlight the feature of the low-contrast defect boundary. A multi-attentional pyramid feature refinement (MPFR) module is then established to refine and fuse the multi-scale defect features of random sizes. Afterwards, a novel global dynamic receptive fusion-transformer (GDRF-Transformer) detection scheme is designed to perform the global perception and feature dynamic extraction of complex internal casting defects. It includes 4D-anchor query and cross-layer box update strategy, query rectification by prior information of defect aspect ratio, and global adaptive-feed forward network (GA-FFN). A dataset comprising turbine blade casting defect radiographic (TBCDR) images, is used to demonstrate the high efficiency of the proposed AGD-DETR. The obtained results show that the proposed method can accurately capture the spatial position distributions and complex defect shapes. Furthermore, it outperforms existing state-of-the-art defect detection methods.

The benefits of Digital Twins are widely recognized across various use cases. However, to ensure efficient utilization of Digital Twins, it is crucial to minimize the effort required for their creation. This is particularly relevant for behavior models, which play a significant role in many Digital Twin use cases. While there are existing approaches for creating these models efficiently, they rely on having access to the asset's structure in a digitally usable format. This requirement also applies to the field of fluidics. The paper presents a method for the automated digitalization of information from fluid circuit diagrams, which contain information about the fluid structure of the asset. The method is implemented on the example of pneumatic vacuum ejectors, and using the test data set as an example, a large part of the information could be digitalized fully automatically. This was also demonstrated for an exemplary circuit diagram with poorer image quality.

Magnetic pulse crimping (MPC) addresses the limitations of conventional hydraulic crimping in cable joint applications. However, the lack of dependable detection methods presents a significant challenge in MPC manufacturing. This study proposed a novel approach integrating 3D vision and ensemble learning to achieve a non-destructive quality assessment of MPC joints. By analyzing the geometric characteristics of crimping products, a specialized 3D vision algorithm was devised to extract geometric features. The random sample consensus (RANSAC) ensured low measurement errors: 0.5 % for terminals and 1.1 % for cables. Coordinate transformation simplified the feature calculation, resulting in an 18.6 % improvement in computational efficiency. To enhance dataset quality, a preprocessing pipeline was designed, incorporating correlation analysis, boxplots, principal component analysis (PCA), and density-based spatial clustering of applications with noise (DBSCAN). It handled irrelevant, redundant, and outlier information effectively. Compared to the original dataset, the training mean squared error (MSE) decreased from 1.790 to 0.290. Additionally, four high-accuracy candidate models were identified via thorough model selection and hyperparameter fine-tuning. Among them, for the design challenge of multilayer perceptron (MLP), a strategy was developed to find an optimal architecture, resulting in a configuration of 3 hidden layers with 16 nodes each. This strategy reduced design variability by constraining hidden layers and ensured stable gradient updates through full-batch training. The candidate models were further integrated using ensemble learning, specifically stacking. The final model achieved a mean absolute error (MAE) of 0.348 kN, and its mean absolute percentage error (MAPE) was 5 %, demonstrating higher accuracy. The results demonstrate the significant potential of the proposed approach in crimping quality prediction, enhancing manufacturing efficiency and reliability.

A Dimensional Variational Prototypical Network (DVPN) is proposed to learn transferable knowledge from a largescale dataset containing sufficient samples of diverse faults, enabling few-shot diagnosis on new faults that are unseen in the dataset. The network includes a multiscale feature fusion module with shared weights to extract fault features, followed by a dimensional variational prototypical module that uses variational inference to determine metric scaling parameters. This adaptive approach accurately measures feature similarity between samples and fault prototypes. To enhance discriminability, a representation learning loss is employed, distinguishing between the least similar samples within the same class (hard positive samples) and the most similar samples across different classes (hard negative samples). The network combines representation learning and prototypical learning through the joint representation learning (JRL) module, acquiring both task-level and feature-level knowledge for a more discriminative metric space and improved classification accuracy on unseen faults. Experimental evaluations on datasets from the Tennessee Eastman process and a real-world polyester esterification process show that the proposed DVPN achieves high diagnostic performance and is comparable to state-of-the-art methods for few-shot fault diagnosis (FSFD).

With the transformation and upgrading of the manufacturing industry, manufacturing systems have become increasingly complex in terms of the structural functionality, process flows, control systems, and performance assessment criteria. Digital representation, performance-related process monitoring, process regulation and control, and comprehensive performance optimization have been viewed as the core competence for future growth. Relevant topics have attracted significant attention and long-term exploration in both the academic and industrial communities. In this paper, focusing on the latest achievements in the context of smart manufacturing, a new performance-driven closed-loop process optimization and control framework with the cloud-edge-device collaboration is proposed. Firstly, in order to fully report the performance optimization and control technologies in manufacturing systems, a comprehensive review of associated topics, including digital representation and information fusion, performance-related process monitoring, dynamic scheduling, and closed-loop control and optimization are provided. Secondly, potential architectures integrating such technologies in manufacturing processes are investigated, and several existing research gaps are summarized. Thirdly, aiming at the hierarchical performance target, we present a roadmap to the cloud-edge-device collaborative closed-loop performance optimization and control for smart manufacturing. The overall architecture, development and deployment, and key technologies are discussed and explored with an actual industrial process scenario. Finally, the challenges and future research focuses are introduced. Through this work, it is hoped to provide new perspectives for the comprehensive performance optimization and control in the transition from Industry 4.0–5.0.