Type-B aortic dissection (TBAD) is a life-threatening vascular condition that requires accurate segmentation of the true and false lumens in computed tomography angiography (CTA) images for precise diagnosis and effective surgical planning. Achieving such segmentation in semi-supervised settings remains challenging, as existing methods often experience performance degradation due to distributional shifts between labeled and unlabeled data domains. To address this issue, this study introduces FMTW-SAM, a novel cross-domain semi-supervised segmentation framework based on the Segment Anything Model (SAM). FMTW-SAM mitigates domain discrepancies through bidirectional foreground copy-paste and enhances pseudo-label quality using a dual-student Mean Teacher network with temporally weighted fusion. Additionally, a foreground-mixing fine-tuning strategy further refines SAM’s understanding of anatomical structures. Experimental results on two TBAD datasets with limited annotations (10% and 20%) demonstrate that FMTW-SAM consistently outperforms state-of-the-art semi-supervised methods and also achieves better performance than the strong fully supervised reference models under the current experimental protocol across multiple evaluation metrics. These results suggest that FMTW-SAM provides an effective solution for cross-domain semi-supervised TBAD segmentation in limited-label settings.

To address the challenges of flow-induced vibration monitoring and prevention in typical industrial pipe-lines,a method for predicting fatigue life based on the equivalent power spectral density is proposed.First,a fatigue monitoring model is established,and a pipeline vibration test bench is constructed through simulation.Using experimental results as examples,fatigue life estimates are calculated using both the time-domain rain-flow counting method and frequency-domain methods,and their performances are compared.The results show that the frequency-domain method based on equivalent acceleration power spectral density effectively reduces spectral leak-age and provides better prediction of fatigue life for pipelines than traditional frequency-domain approaches.On average,the prediction accuracy improves by 3.53％,and the mean prediction error is the lowest.This study offers an effective approach for predicting the remaining fatigue life of industrial pipelines and offers a novel method for fatigue damage assessment.

In mixed domain semi-supervised medical image segmentation (MiDSS), achieving superior performance under domain shift and limited annotations is challenging. This scenario presents two primary issues: (1) distributional differences between labeled and unlabeled data hinder effective knowledge transfer, and (2) inefficient learning from unlabeled data causes severe confirmation bias. In this paper, we propose the bidirectional correlation maps domain adaptation (BCMDA) framework to overcome these issues. On the one hand, we employ knowledge transfer via virtual domain bridging (KTVDB) to facilitate cross-domain learning. First, to construct a distribution-aligned virtual domain, we leverage bidirectional correlation maps between labeled and unlabeled data to synthesize both labeled and unlabeled images, which are then mixed with the original images to generate virtual images using two strategies, a fixed ratio and a progressive dynamic MixUp. Next, dual bidirectional CutMix is used to enable initial knowledge transfer within the fixed virtual domain and gradual knowledge transfer from the dynamically transitioning labeled domain to the real unlabeled domains. On the other hand, to alleviate confirmation bias, we adopt prototypical alignment and pseudo label correction (PAPLC), which utilizes learnable prototype cosine similarity classifiers for bidirectional prototype alignment between the virtual and real domains, yielding smoother and more compact feature representations. Finally, we use prototypical pseudo label correction to generate more reliable pseudo labels. Empirical evaluations on three public multi-domain datasets demonstrate the superiority of our method, particularly showing excellent performance even with very limited labeled samples. Code available at https://github.com/pascalcpp/BCMDA.

Accurate aerodynamic characterization is crucial for optimizing aircraft design and enhancing flight performance. Multi-fidelity modeling approaches improve aerodynamic prediction accuracy and computational efficiency by integrating data from various fidelity levels. To better handle the complex mixed linear and nonlinear correlations coexisting between high- and low-fidelity data, this paper proposes a new multi-fidelity Gaussian process regression (MFGPR) model based on the nonlinear autoregressive Gaussian process (NARGP) framework. By integrating linear and nonlinear kernel functions, the proposed model extends the capabilities of NARGP, enabling it to simultaneously capture complex nonlinear relationships and linear dependencies within multi-fidelity data. To validate the effectiveness of the MFGPR method, two classes of classic analytical functions were selected for numerical testing, and a comparative analysis was performed against three traditional multi-fidelity methods: Cokriging, NARGP, and MFDNN. The results indicate that in handling linear correlations, the prediction performance of MFGPR is consistent with that of CoKriging. Conversely, in modeling nonlinear correlations, MFGPR demonstrates higher prediction accuracy than the other three methods, while offering a clear advantage in modeling efficiency. Furthermore, MFGPR was applied to predict the pressure distribution of the ONERA M6 wing and the drag coefficient of the NACA2414 airfoil, verifying its potential application and superior performance in aerodynamic modeling.

Aortic Dissection (AD) is a life-threatening disease that can be rapidly screened by using deep learning methods. However, deep learning model training requires a large amount of manual annotation of data. To improve the annotation efficiency and accuracy, we propose LABELSAM, a semi -automatic interactive segmentation algorithm designed to efficiently annotate AD in 3D computed tomography angiography (CTA) images. By requiring minimal user input only points and bounding boxes on the first and last slices LABEL-SAM automatically generates segmentation prompts for intermediate slices, reducing the manual workload. In addition, we propose a bidirectional prediction weighting method and a fine-tuning strategy tailored to AD data, further enhancing segmentation accuracy. Furthermore, LABEL-SAM is implemented as a plug-and-play plugin for the 3D Slicer software. Experimental results on both external and internal datasets demonstrate the method's superior performance, improving annotation accuracy and efficiency. The code will be available at htips://github.comNenjiecai/LABEL-SAM. The demonstration video is now available at htips://www.3oulube.rom/waleh? 31 zg11b4JQ.

Using deep learning to reconstruct CT from X-ray images not only overcomes the inherent 2D limitations of X-ray imaging but also reduces radiation exposure and costs associated with CT scans. While current deep learning approaches offer feasible solutions, they often overlook the role of attention mechanisms in capturing critical anatomical features. Moreover, these methods ignore the perceptual consistency between the reconstructed CT images and the ground truth in terms of luminance, contrast, and structure. To address these limitations, we propose a novel medical image reconstruction framework, namely OX2CT-GAN, which aims to reconstruct 3D chest CT images from 2D orthogonal biplanar X-ray images. Specifically, we propose Feature Enhancement Blocks (FEB), which are embedded in the encoding stage of the generator to enhance the model’s ability to extract critical anatomical features. Additionally, we propose a Perceptual Consistency Loss to enhance the quality of the reconstructed CT images. Experimental results demonstrate that our method significantly outperforms existing deep learning models in both quantitative metrics and qualitative evaluations, achieving state-of-the-art (SOTA) performance with promising applications in the field.

Early treatment of pancreatic cancer is critical for improving patient survival rates. Accurate segmentation of the pancreas and its tumors in CT images is essential for developing effective clinical strategies. However, this task remains highly challenging due to the small size, low contrast and significant anatomical variability of pancreatic tumors. To overcome these challenges, DMSD-Net is introduced as an innovative segmentation model built upon nnU-Net. It consists of two key components: the DMSDA Block and the DMSD Attention Mechanism. The DMSDA Block adjusts local features with global context, while the DMSD Attention Mechanism adaptively modifies the receptive field, improving the model's capability to detect intricate and irregular targets. This architecture effectively combines global and local information, enhancing the model's ability to capture fine details in medical images. We evaluated DMSD-Net on two public datasets, MSD and NIH. The results demonstrate that our model surpasses current leading segmentation methods. Furthermore, ablation studies confirm the individual contributions of the DMSDA Block and DMSD Attention Mechanism. In conclusion, our study shows that DMSD-Net supports the diagnosis and preoperative planning of pancreatic cancer, laying a foundation for future research and clinical use.

Considering the problems of having insufficient fault identification from single information sources in actual industrial environments, and different information sensitivity in multi-information source data, and different sensitivity of artificial feature extraction, which can lead to difficulties of effective fusion of equipment information, insufficient state representation ability, low fault identification accuracy, and poor robustness, a multi-information fusion fault identification network model based on deep ensemble learning is proposed. The network is composed of multiple sub-feature extraction units and feature fusion units. Firstly, the fault feature mapping information of each information source is extracted and stored in different sub-models, and then, the features of each sub-model are fused by the feature fusion unit. Finally, the fault recognition results are obtained. The effectiveness of the proposed method is evaluated by using two gearbox datasets. Compared with the method of simple stacking fusion and single measuring point without fusion, the accuracy of each type of fault recognition of the proposed method is close to 100%. The results show that the proposed method is feasible and effective in the application of gearbox fault recognition.

For computational fluid dynamics (CFD) numerical simulations of complex geometries, mesh generation faces numerous challenges, especially in generating high-quality anisotropic meshes in regions with complex boundaries. Selecting an appropriate topological structure is a key difficulty in the process of anisotropic mesh generation. Traditional methods for generating anisotropic mesh topologies mainly rely on heuristic ideas and prior knowledge, lacking theoretical support. This easily leads to problems such as poor mesh adaptability, intersection of mesh lines, and local distortion, which in turn limits the stability and quality of mesh generation. To address this bottleneck, based on the Poincaré conjecture and the Helmholtz theorem, this paper proposes a streamline-based topological structure, which is applied to mesh generation for complex geometries in CFD. This method introduces the characteristic of non-intersecting streamlines into the mesh generation process, effectively avoiding the problems of self-intersection of mesh lines and local distortion both theoretically and practically, and significantly improving the adaptive mesh generation ability. The experimental results show that when dealing with complex geometries, the proposed method not only significantly improves the mesh quality but also enhances the stability and reliability of the mesh generation process, demonstrating advantages that are incomparable to those of traditional methods.

Bearing health state recognition is often affected by variable speeds and heavy load working conditions, making fault signal features difficult to identify and resulting in challenges in health condition recognition methods, including feature extraction difficulties and cross-device recognition challenges. This paper proposes the Bearing Health Status Evaluation Method based on Multi-Scale Hybrid Features and Inception-Block Attention Bidirectional Physics-Informed Domain Adaptation Network. A multi-scale feature extraction method is developed, and a Multi-Scale Hybrid Features and Inception-Block Attention Bidirectional Physics-Informed Domain Adaptation Network is introduced. This network uses physical-information layers and inverse-physical-information layers to constrain multi-scale features and adaptively adjust model hyperparameters, incorporating Inception multi-scale convolution and convolutional self-attention mechanisms to enhance feature recognition capabilities. To validate the effectiveness of the model, this paper constructs a Health Status Dynamic Time Warping-Mic Index and uses the Xi’an Jiao tong University bearing degradation dataset, PHM2012 challenge dataset, and centrifugal pump engineering data for model validation. The results demonstrate that the model performs well in recognizing the health status of equipment under cross-operating conditions and cross-device scenarios.

X-rays are widely used in clinical practice due to their low radiation exposure and cost. However, 2D imaging can result in overlapping anatomical structures. In contrast, CT scans generate 3D images, effectively addressing this limitation. Nevertheless, CT scans also have drawbacks, such as high radiation, high cost, and inability to be implemented within ICU settings. In this paper, we propose the AP2CT-GAN framework, which aims to reconstruct chest CT images from a single antero-posterior chest X-ray. The framework incorporates the Feature Enhancement Connection (FEC) and the Feature Dimension Converter (FDC) to enhance critical features and capture global contextual information, along with the Dual-Consistency Loss function propose to ensure that the reconstructed CT images maintain a high level of structural and textural consistency with the ground truth. Experimental results demonstrate that AP2CT-GAN outperforms existing methods, offering a low-radiation, low-cost CT imaging solution with valuable potential applications in resource-limited regions and ICU settings.

Graph Neural Networks (GNNs) commonly employ sampling-based methods for inference on large-scale real-world graphs. However, the inherent characteristics of sampling lead to redundant data loading during GNN inference, while slow data transfer between the host and GPU exacerbates the issues of slow inference and low resource utilization. Current methods to accelerate GNN inference face several challenges: (1) low GPU resource utilization; (2) neglect of adjacency matrix locality; and (3) long preprocessing time. To address these issues, we propose DCI, a system designed to accelerate GNN inference. The system provides a simple and effective cache capacity allocation and filling strategy that can adapt flexibly to different workload demands. During the pre-sampling phase, DCI allocates and fills cache capacities for node features and adjacency matrices based on workload patterns. Experimental results show that DCI accelerates sampling and node feature loading, achieving end-to-end inference speedups of 1.18 × to 11.26 × compared to DGL, and 1.14 × to 13.68 × compared to RAIN, while reducing preprocessing time by 52.8 × to 1.32 × . We also compared DCI with DUCATI’s dual-cache population strategy, and DCI achieves nearly identical inference speeds while reducing preprocessing time to less than 20

In response to the issues of poor noise resistance, difficulty in extracting faint features, and confusion of signal characteristics across different fault statuses observed in existing entropy features for cross-domain diagnosis of rolling bearings, a refined composite zoom multiscale weighted permutation entropy (RCZMWPE) index has been proposed. This index is capable of extracting faint fault signals across the entire frequency band. Furthermore, a diagnostic model based on RCZMWPE, incorporating ensemble feature learning and semisupervised manifold feature transfer for cross-domain fault diagnosis, has been constructed. First, by decomposing the raw vibration signals, effective components are selected for reconstruction and enhancement to reinforce the expression of sensitive fault features. Second, an ensemble feature learning classification strategy based on multikernel twin underlying classifiers is proposed, which continuously enhances the learning capability of the weak classifiers for common features and integrates them into a strong classifier. Then, the dynamic distribution differences of cross-domain feature samples are measured, and the manifold feature transfer strategy is utilized to optimally map and identify the health status of bearings. Finally, the effectiveness and robustness of the proposed method are demonstrated through four public datasets of different test benches, four sets of real engineering case data from different equipment, and aeroengine bearing data with different degrees of fault damage. Additionally, a comparison with seven published entropy features and six fault diagnosis methods from references demonstrates that the proposed method exhibits superior performance in handling multisource cross-domain diagnostic tasks across different equipment and operating conditions.

Traditional super-resolution reconstruction methods for flow fields use end-to-end mapping to determine the relationship between high- and low-resolution flow field data. The reconstruction quality of these methods depends on the accuracy of the low-resolution data. Ensuring the accuracy of low-resolution data has, thus, become a precondition for super-resolution tasks, and it imposes strict limitations on the applicability of super-resolution reconstruction methods in practical engineering applications. This paper proposes a flow field super-resolution reconstruction method coupled with feature recognition (FRNet) to reduce the dependence on the accuracy of low-resolution data. FRNet uses a feature extractor with identification capabilities to determine the effectiveness of low-resolution flow field characteristics. It recognizes the effective characteristics using a feature distance distribution. Meanwhile, a representation of the obstacle shape and freestream information is introduced to compensate for invalid features and to suppress the influence of low-precision flow field characteristics on the reconstruction results. Different downsampling factors, different density grids, and noise are used to simulate a variety of engineering application scenarios to verify the effectiveness and applicability of the proposed method. The results demonstrate that FRNet has significant advantages over traditional super-resolution reconstruction methods. Our method does not rely on the accuracy of low-resolution data and can effectively mitigate the impact of low-resolution flow field data that do not conform to physical phenomena. This characteristic allows FRNet to exhibit outstanding performance when handling flow field data affected by noise from wind tunnel wall and rack interferences. Consequently, FRNet should prove highly beneficial for the optimization of complex flow fields using super-resolution reconstruction methods.

Early fault warning for large-scale high-speed rotating machinery can effectively reduce unplanned downtime and avoid major safety accidents. Aiming at the problems of difficult screening of multi-source common sensitive features, the challenging training of neural networks with a small number of sensitive features, and the difficulty of directly using generative adversarial networks for early fault warning, this paper constructs an early fault warning model based on multi-source common sensitive features and an improved Wasserstein generative adversarial network, proposing an early fault warning method for rotating machinery. The model was verified by using the open XJTU-SY bearing laboratory data, the P3409A centrifugal pump bearing fault engineering case data of a petrochemical company and the rotor system engineering case data of a circulating hydrogen centrifugal compressor of a petrochemical company. The early fault warning method of rotating machinery proposed in this paper warns the bearing fault of centrifugal pump 160 hours in advance and the rotor system fault of centrifugal compressor 1330 minutes in advance. Compared with the two published methods, the proposed method has better early fault warning effect, better normal and abnormal health index discrimination and less false warning.

The introduction of deep learning has resolved the high-cost issues associated with traditional methods in handling complex aerodynamics problems and is commonly used for simulating fluid behavior and optimizing aircraft design. However, flow field prediction based on deep learning typically encodes the freestream conditions and geometric information into the neural network model concurrently. This encoding scheme makes it difficult for the model to distinguish and deal with the intrinsic differences between these two types of information. As a result, the ability of the model to capture complex flow field features decreases and the difficulty of model fitting increases, which in turn reduces the effectiveness of the model. To solve these problems, this paper proposes the Operator-Convolution MultiModal Fusion Network (OCMMFNet), a new neural network architecture to predict the flow fields of airfoils with various geometries and freestream conditions. The proposed network architecture uses a freestream generalization network to encode the input freestream conditions. The resulting approximate flow field information is combined with the airfoil geometry information and fed into a shape feature compensation network to improve the prediction accuracy. We compare the performance of OCMMFNet with those of a deep operator network(DeepONet) and a vision transformer(ViT) model. When generalizing both freestream conditions and airfoil shapes, OCMMFNet reduces the prediction error in the pressure field by 9.71% and 3.76% compared to DeepONet and ViT, respectively. In tests involving extrapolation of Reynolds numbers, OCMMFNet significantly reduces the prediction error in the pressure field by 13.73% and 11.84% compared to DeepONet and ViT, respectively. The results show that OCMMFNet achieves better prediction accuracy than both DeepONet and ViT and displays superior robustness and generalization ability.

Aortic Dissection (AD) is a life-threatening disease. As such, it’s vital to make clinical strategies by accurate segmentation of AD from Computed Tomography Angiography (CTA) images. Due to the heterogeneous morphology of AD in CTA images and the influence of image noise, inaccurate segmentation results are often produced, making it challenging to accurately identify the contours of AD. To meet these challenges, we propose a novel automatic AD segmentation model DETR-SAM based on the Segment Anything Model(SAM). Our model consists of a deformable DETR object detector module with multi-scale feature fusion and a fine-tuned SAM module for refining the segmentation contours. The object detector module effectively fuses the multi-scale image features extracted by the image encoder of SAM with the features obtained from the DETR backbone CNN network. The fused multi-scale features generate multiple prompts rich in geometric shape knowledge through multiple regression prediction heads to guide the SAM model to refine the segmentation boundaries and enhance the recognition of contours. Meanwhile, a fine-tuning strategy using these prompts is proposed to enhance the SAM model’s understanding of AD features. This strategy enables our DETR-SAM model with a 2D segmentation network architecture to surpass the performance of mainstream 3D segmentation algorithms. We have evaluated DETR-SAM on the ImageTBAD dataset and our internal dataset. The results showed that the accuracy of our model is better than the current state-of-the-art 2D and 3D segmentation methods in all four indicators. In addition, the effectiveness of the feature fusion module, hint generation method and fine-tuning strategy are further verified by ablation experiments. Our DETR-SAM method is helpful to a certain extent in the diagnosis and auxiliary treatment of AD in different clinical situations and lays a foundation for further research and clinical application.

To address the lack of reliable measurement methods for identifying wear mechanisms and predicting the state of mechanical seal tribo-parts, this study proposes a method for characterizing tribological behavior based on measuring face vibration acceleration. It aims to uncover the source mechanism of mechanical seal face vibration acceleration influenced by tribology and dynamic behavior. This research delves into the dynamic behavior characteristics and vibration acceleration of the mechanical seal stationary ring. We explored the variation pattern of face vibration acceleration root mean square (RMS) with rotation speed, sealing medium pressure, and face surface roughness. The results indicate that under constant medium pressure, an increase in rotation speed leads to a decrease in acceleration RMS and an increase in face temperature. Similarly, under constant rotation speed, an increase in medium pressure results in nonlinear changes in acceleration RMS, forming an “M” shape, along with an increase in face temperature. Furthermore, under conditions of constant medium pressure and rotation speed, an increase in the surface roughness of the rotating ring face corresponds to an increase in acceleration RMS and face temperature. Upon starting the mechanical seal, both acceleration RMS and temperature initially increase before decreasing, a trend consistent with the Stribeck curve.

At present, high-fidelity data are expensive to acquire. When fusing limited high-fidelity data, the small-sample size introduces problems such as missing information and sample bias, which leads to overfitting of the results and accuracy degradation. In this paper, we propose a small-sample aerodynamic data fusion method based on deep neural networks. The method applies semi-supervised learning for model construction using multi-fidelity aerodynamic thermal and force data. The initial model is trained with both labeled and unlabeled data by an improved flexible loss function. Using unlabeled data as a soft constraint combined with semi-supervised learning enables the model to perform better with small-sample data. This article investigates the ONERA (National Office for Aerospace Studies and Research) M6 wing surface pressure distributions at different airfoil spread coordinates and verifies the applicability of the proposed method by reducing the proportion of high-fidelity data in the training and test datasets. The proposed method is then applied to the prediction of aerothermal data on the surface of a blunt bicone. The results show that, using a small-sample high-fidelity dataset, the proposed method can predict the surface pressure distribution and surface aerodynamic heat distribution of the aircraft relatively well. As the volume of high-fidelity data decreases, the proposed method outperforms other methods.

To enable e-commerce sellers to identify products favoured by consumer groups and thereby increase their store's sales and profit margins, a product classification model based on feature-level fusion of heterogeneous data is proposed. The model uses a convolutional neural network to learn feature representations from product title, descriptions and reviews. It also uses a deep autoencoder to learn feature representations from product attributes. By combining the features of heterogeneous data, a comprehensive representation of the product is obtained. Experiments were conducted using data from all basketball products on the Amazon e-commerce platform. The results of the experiments showed that building a product classification model based on heterogeneous data outperformed using only product title. The accuracy of the model improved by 3.13% and the F1 score increased by 4.54%. These results strongly suggest that the product classification model based on heterogeneous data has learned more comprehensive features related to product popularity.