Two-stage Framework Research Articles

A two-stage uncertainty quantification framework for reliability and sensitivity analysis of structures using the probability density evolution method integrated with the Fréchet-derivative-based method

In this work, a novel Fréchet-derivative-based global sensitivity analysis is incorporated into the probability density evolution method for the simultaneous reliability and sensitivity analysis of structures. The two methods are synthesized and realized under a two-stage uncertainty quantification framework. Fundamental and numerical algorithms of the method are described, which is validated by a benchmark problem with theoretical solutions. Then it is applied on studying stochastic seismic responses of a real-world reinforced concrete structure of 288 m high. Results demonstrate that the response characteristics of the structure will be greatly changed when both uncertainties from the ground motions and structural parameters are considered, e.g., the failure probability will be enlarged by 1∼2 orders of magnitude. Besides, contrary to the popular opinion that the uncertainty of structural parameters is trivial compared to that of parameters in the ground-motion model, the importance measures of the structural parameters can be greater than those of the parameters in the ground-motion model by a factor of 2. when the coupling effect of randomness and nonlinearity of structural materials becomes dramatic. Some issues to be further outlined are also discussed.

Self-supervised deep learning for joint 3D low-dose PET/CT image denoising

Deep learning (DL)-based denoising of low-dose positron emission tomography (LDPET) and low-dose computed tomography (LDCT) has been widely explored. However, previous methods have focused only on single modality denoising, neglecting the possibility of simultaneously denoising LDPET and LDCT using only one neural network, i.e., joint LDPET/LDCT denoising. Moreover, DL-based denoising methods generally require plenty of well-aligned LD-normal-dose (LD-ND) sample pairs, which can be difficult to obtain. To this end, we propose a self-supervised two-stage training framework named MAsk-then-Cycle (MAC), to achieve self-supervised joint LDPET/LDCT denoising. The first stage of MAC is masked autoencoder (MAE)-based pre-training and the second stage is self-supervised denoising training. Specifically, we propose a self-supervised denoising strategy named cycle self-recombination (CSR), which enables denoising without well-aligned sample pairs. Unlike other methods that treat noise as a homogeneous whole, CSR disentangles noise into signal-dependent and independent noises. This is more in line with the actual imaging process and allows for flexible recombination of noises and signals to generate new samples. These new samples contain implicit constraints that can improve the network’s denoising ability. Based on these constraints, we design multiple loss functions to enable self-supervised training. Then we design a CSR-based denoising network to achieve joint 3D LDPET/LDCT denoising. Existing self-supervised methods generally lack pixel-level constraints on networks, which can easily lead to additional artifacts. Before denoising training, we perform MAE-based pre-training to indirectly impose pixel-level constraints on networks. Experiments on an LDPET/LDCT dataset demonstrate its superiority over existing methods. Our method is the first self-supervised joint LDPET/LDCT denoising method. It does not require any prior assumptions and is therefore more robust.

Target Soybean Leaf Segmentation Model Based on Leaf Localization and Guided Segmentation

The phenotypic characteristics of soybean leaves are of great significance for studying the growth status, physiological traits, and response to the environment of soybeans. The segmentation model for soybean leaves plays a crucial role in morphological analysis. However, current baseline segmentation models are unable to accurately segment leaves in soybean leaf images due to issues like leaf overlap. In this paper, we propose a target leaf segmentation model based on leaf localization and guided segmentation. The segmentation model adopts a two-stage segmentation framework. The first stage involves leaf detection and target leaf localization. Based on the idea that a target leaf is close to the center of the image and has a relatively large area, we propose a target leaf localization algorithm. We also design an experimental scheme to provide optimal localization parameters to ensure precise target leaf localization. The second stage utilizes the target leaf localization information obtained from the first stage to guide the segmentation of the target leaf. To reduce the dependency of the segmentation results on the localization information, we propose a solution called guidance offset strategy to improve segmentation accuracy. We design multiple guided model experiments and select the one with the highest segmentation accuracy. Experimental results demonstrate that the proposed model exhibits strong segmentation capabilities, with the highest average precision (AP) and average recall (AR) reaching 0.976 and 0.981, respectively. We also compare our segmentation results with current baseline segmentation models, and multiple quantitative indicators and qualitative analysis indicate that our segmentation results are better.

Combination of top-down and bottom-up DEA models using PCA: A two-stage network DEA with shared input and undesirable output for evaluation of the road transport sector

This paper presents a network data envelopment analysis (NDEA) model to assess road transport sector by considering energy and non-energy inputs, shared input, and desirable and undesirable outputs. The sector is considered as a two-stage framework where in stage 1, energy-environmental efficiency is calculated and in stage 2, technical efficiency is estimated. There are two general approaches for evaluation of the network structures using NDEA models: bottom-up and top-down. In the bottom-up approach of this study, a goal programming NDEA (GP-NDEA) model is proposed to calculate the efficiency of the two stages at the same time and then overall efficiency is obtained using the results of the two stages. In the top-down approach, both non-radial SBM-NDEA and RAM-NDEA models are applied to calculate the overall efficiency and then the efficiencies of stage 1 and stage 2 are estimated using the results of the overall efficiency. Three different models GP-NDEA, SBM-NDEA, and RAM-NDEA lead to different results and to obtain the final efficiencies and ranks, principal component analysis (PCA) method is applied. The efficiency generated by different bottom-up and top-down approaches are considered as indicators in PCA, and the final scores for both stages are calculated using PCA. To illustrate the capability of the proposed approach, an actual dataset of road transport sector in provinces of Iran is gathered and the results of the different models are analyzed.

An integrated decision support framework for resilient vaccine supply chain network design

Designing resilient supply chain networks for vaccine development and distribution requires reliable and robust infrastructure. This stud develops a novel two-stage decision support framework for configuring multi-echelon Supply Chain Networks (SCNs), resilient supplier selection, and order allocation under uncertainty. Resilient supplier selection is done using a hybrid Multi-Criteria Decision-Making (MCDM) approach based on Best-Worst Method (BWM), Weighted Aggregated Sum Product Assessment (WASPAS), and Type-2 Neutrosophic Fuzzy Numbers (T2NN). A robust multi-objective optimization model is then built for order allocation considering resiliency scores, reliability of facilities, and uncertain supply and demand. The objectives are to minimize the total cost of SCN design, maximize the resiliency score, and maximize the reliability of SC, respectively. A Non-dominated Sorting Genetic Algorithm II (NSGA-II) is developed to tackle the problem on large scales, tuned by the Taguchi design technique. The NSGA-II solution is compared to the ε-constraint and Multi-objective Particle Swarm Optimization (MOPSO) solutions using test problems. We demonstrate the superiority of the suggested NSGA-II method over the two competing methods according to five performance metrics. A case study is then investigated to illustrate the applicability and effectiveness of the offered methodology for COVID-19 vaccine distribution in a developing country. It is revealed that the models and algorithms can treat the problem optimally, such that Germany is the main source (approximately 25.61%) while India does not contribute to the supply of vaccines.

Two-stage contextual transformer-based convolutional neural network for airway extraction from CT images

Accurate airway segmentation from computed tomography (CT) images is critical for planning navigation bronchoscopy and realizing a quantitative assessment of airway-related chronic obstructive pulmonary disease (COPD). Existing methods face difficulty in airway segmentation, particularly for the small branches of the airway. These difficulties arise due to the constraints of limited labeling and failure to meet clinical use requirements in COPD. We propose a two-stage framework with a novel 3D contextual transformer for segmenting the overall airway and small airway branches using CT images. The method consists of two training stages sharing the same modified 3D U-Net network. The novel 3D contextual transformer block is integrated into both the encoder and decoder path of the network to effectively capture contextual and long-range information. In the first training stage, the proposed network segments the overall airway with the overall airway mask. To improve the performance of the segmentation result, we generate the intrapulmonary airway branch label, and train the network to focus on producing small airway branches in the second training stage. Extensive experiments were performed on in-house and multiple public datasets. Quantitative and qualitative analyses demonstrate that our proposed method extracts significantly more branches and longer lengths of the airway tree while accomplishing state-of-the-art airway segmentation performance. The code is available at https://github.com/zhaozsq/airway_segmentation.

An Integrated Approach Based on MADM and MODM for Order Picking System Considering Human Factors

Optimizing order-picking systems (OPSs) while considering human factors and integrating key decisions is a major challenge for warehouse managers. This study presents a two-stage framework based on multi-attribute decision-making (MADM) and multi-objective decision-making (MODM) models to integrate decisions on picker selection, order batching, batch assignment, picker routing, and scheduling. In the first stage, the human factors affecting picker selection are considered as the problem’s criteria and the available pickers are treated as alternatives. The fuzzy entropy method and fuzzy COmplex PRoportional ASsessment (COPRAS) are used to weight the factors and rank the pickers, respectively. In the second stage, a three-objective mathematical model is formulated to minimize makespan and the operating costs of picking while maximizing the total scores of the selected pickers. The improved augmented epsilon constraint method (AUGMECON2) and the non-dominated sorting genetic algorithm II (NSGA-II) are applied to solve the proposed model. The performance of the two methods is tested on well-known benchmark instances and a real-world case study. The NSGA-II algorithm can generate optimal results using only about 6.58% of the CPU time required by AUGMECON2 to solve the problem. Our computational experiments show that increasing the number of pickers from 2 to 8 and doubling their capacity reduces the makespan by 2.61% and 2.74%, respectively.

Two-stage automated diagnosis framework for urogenital schistosomiasis in microscopy images from low-resource settings.

Automated diagnosis of urogenital schistosomiasis using digital microscopy images of urine slides is an essential step toward the elimination of schistosomiasis as a disease of public health concern in Sub-Saharan African countries. We create a robust image dataset of urine samples obtained from field settings and develop a two-stage diagnosis framework for urogenital schistosomiasis. Urine samples obtained from field settings were captured using the Schistoscope device, and S. haematobium eggs present in the images were manually annotated by experts to create the SH dataset. Next, we develop a two-stage diagnosis framework, which consists of semantic segmentation of S. haematobium eggs using the DeepLabv3-MobileNetV3 deep convolutional neural network and a refined segmentation step using ellipse fitting approach to approximate the eggs with an automatically determined number of ellipses. We defined two linear inequality constraints as a function of the overlap coefficient and area of a fitted ellipses. False positive diagnosis resulting from over-segmentation was further minimized using these constraints. We evaluated the performance of our framework on 7605 images from 65 independent urine samples collected from field settings in Nigeria, by deploying our algorithm on an Edge AI system consisting of Raspberry Pi + Coral USB accelerator. The SH dataset contains 12,051 images from 103 independent urine samples and the developed urogenital schistosomiasis diagnosis framework achieved clinical sensitivity, specificity, and precision of 93.8%, 93.9%, and 93.8%, respectively, using results from an experienced microscopist as reference. Our detection framework is a promising tool for the diagnosis of urogenital schistosomiasis as our results meet the World Health Organization target product profile requirements for monitoring and evaluation of schistosomiasis control programs.

Location and inventory pre-positioning problem under uncertainty

In humanitarian logistics, the location and inventory pre-positioning problem (LIPP), making decisions on facility location, inventory pre-positioning, and allocation, is critical to post-disaster rescue efficiency. Since post-disaster conditions are complex and uncertain, it is challenging for planners to make LIPP decisions and perform rescue activities. To reduce operational costs while also guaranteeing service levels to customers, it is necessary to take uncertainties into account when making decisions. This study explores a service-oriented LIPP under two types of uncertainties, i.e., uncertain customer demand and third-party supply. We propose a two-stage robust optimization framework to solve the problem, considering the service-level constraints and minimizing the total cost. We develop two algorithms to solve the problem. Specifically, a column-and-constraint generation algorithm is utilized to solve the original two-stage robust model, and an improved row generation algorithm is developed to tackle the affinely adjustable robust counterpart model. Simulation results show that both algorithms can provide satisfying solutions. A case study based on a power grid company in China is conducted to perform sensitivity analysis and compare models, producing several managerial insights.

Two-stage scheduling of smart electric vehicle charging stations and inverter-based Volt-VAR control using a prediction error-integrated deep reinforcement learning method

Smart electric vehicle charging stations (EVCSs) having distributed energy resources (DERs), including photovoltaic (PV) systems and energy storage systems (ESSs), are becoming vital devices for increasing their profit and maintaining stable distribution grid operations by scheduling the real/reactive power of DERs. However, prediction errors of PV generation outputs and electric vehicle (EV) loads from EVCSs may decrease their profit and destabilize the distribution grid owing to incorrect EV charging scheduling and voltage regulation. To address this issue, we propose a two-stage framework for smart EVCS scheduling and inverter-based Volt-VAR control (VVC) using prediction error-integrated deep reinforcement learning (DRL). In the first stage, the EVCS agents train their neural network model with a 30-min resolution to maximize their profit through a day-ahead charging/discharging scheduling of ESSs in the EVCSs while responding to various prediction errors of PV generation outputs and EV loads. The total real power consumption of each EVCS, including the charging/discharging schedules of the ESSs calculated in the first stage, is delivered to the second stage, in which the VVC agent trains its neural network model with a 5-min resolution to minimize the real power loss and voltage violations through real-time reactive power scheduling of the ESSs in the EVCSs via their inverters. The proposed approach was tested in the IEEE 33-node and IEEE 123-node distribution systems. The results show that the proposed approach outperforms DRL methods that do not consider the prediction errors in terms of profitability of the EVCS and reduction of real power loss.

SEGANet: 3D object detection with shape-enhancement and geometry-aware network

3D object detection approaches from point clouds develop rapidly. However, the distribution of point clouds is unbalanced in the real scene, and thus the distant or occluded objects suffer from too few points to be perceived. This case damages the overall detection accuracy. Hence, we propose a novel two-stage 3D object detection framework, the Shape-Enhancement and Geometry-Aware Network (SEGANet), which aims to mitigate the negative impact of unbalanced point distribution for boosting detection performance. In stage 1, we first capture fine-grained structural knowledge with the assistance of point-wise features from voxels to generate proposals. And in stage 2, we construct a shape enhancement module to reconstruct complete surface points for objects within proposals, then establish an elaborate geometric relevance-aware Transformer module to aggregate high-correlated feature pairs of reconstructed-known parts and decode vital geometric relations of aggregated features. Thus, critical geometric clues are supplied for objects from the data and feature levels, achieving enhanced features for box refinement. Extensive experiments on KITTI and Waymo datasets show that SEGANet achieves low model complexity and excellent detection accuracy, especially leading the baseline method by 2.18% gain in overall detection accuracy and 1.8% gain in average accuracy of weakly sensing objects. This verifies that SEGANet effectively alleviates the impact of point imbalance to significantly boost detection performance.

Identification of disulfidptosis-related subtypes and development of a prognosis model based on stacking framework in renal clear cell carcinoma.

Clear cell renal cell carcinoma (ccRCC) is a common malignant tumor with an unsatisfactory prognosis. This study aims to identify the expression patterns of disulfidptosis-related genes (DRGs), develop a prognostic model, and predict immunological profiles. First, we identified differentially expressed DRGs in TCGA-KIRC cohort and analyzed their mutational profiles, methylation levels, and interaction networks. Subsequently, we identified disulfidptosis-associated molecular subtypes and investigated their prognostic and immunological characteristics. Simultaneously, a disulfidptosis-related prognostic signature (DRPS) was developed using a two-stage stacking framework consisting of 5 machine learning models. The effect of DRPS on immune cell infiltration levels was explored using seven different algorithms, and the status and function of T cells for distinct risk-score groups were evaluated based on T cell exhaustion and dysfunction scores. Additionally, the study also examined differences in clinical characteristics and therapy efficacy between high- and low-risk groups. We found two disulfidptosis-associated clusters, one of which had a poor prognosis and was linked to high immune cell infiltration but impaired T cell function. DRPS showed excellent predictive performance in all four cohorts and could accurately identified disulfidptosis-related molecular subtypes. The DRPS-based risk score was positively associated with poor prognosis, malignant pathological features, high immune cell infiltration levels, and T cell exhaustion or dysfunction, and better respond to immunotherapy and targeted therapy.Additionally, we have identified a close association between ISG20 and disulfidptosis as well as tumor immunity. Our study identified distinct disulfidptosis-related subtypes in ccRCC patients, and constructed the highly accurate and robust DRPS based on an ensemble learning framework, which has critical reference value in clinical decision-making and individualized treatment.And this workalso revealedISG20 exhibits promising potential as a therapeutic target for ccRCC.

Operational and economy-wide impacts of compound cyber-attacks and extreme weather events on electric power networks

The growing frequencies of extreme weather events and cyber-attacks give rise to a novel threat where a malicious cyber actor aims to disrupt stressed components of critical infrastructure systems immediately before, during, or shortly after an extreme weather event. In this paper, we initiate the study of Compound Cyber–Physical Threats and develop a two-stage framework for the analysis of operational disruptions in electric power networks and economy-wide impacts under three scenarios: a Heatwave, a Cyber-attack, and a Compound scenario when the Cyber-attack is timed with the Heatwave. In the first stage, we use a bilevel optimization problem to represent the adversarial rationale of a cyber-attacker in the upper level. In the lower level, we model disruptions in the electric power network using an optimal power flow model. In the second stage, we couple the disruption of electricity supply with a Computable General Equilibrium model to elucidate the impacts on all economic sectors. For the New York Independent System Operator, we find that a 9% demand increase in a Heatwave may not lead to unserved load. The cyber-attack can lead to 4% of unserved electric load in Long Island, while the Compound scenario can increase unserved electric load in Long Island to 12% and affect almost 198,000 customers. Our results show that the activity of state and local government enterprises can decrease by 37% in the Compound scenario. We conclude that the vulnerability of federal, state, and local government enterprises to electricity disruptions can affect a broad range of populations.

Relative Entropy-Based Constant-Envelope Beamforming for Target Detection in Large-Scale MIMO Radar With Low-Resoultion ADCs

Hybrid digital/analog architecture and lowresolution analog-to-digital/digital-to-analog converters (ADCs /DACs) are two low-cost implementations for large-scale millimeter wave (mmWave) systems. In this paper, we investigate the problem of constant-envelope transmit beamforming for large-scale multiple-input multiple-output (MIMO) radar system, where the transmit array adopts a hybrid digital/analog architecture with a small number of RF chains and the receive array adopts a fully digital architecture with low-resolution ADCs. We derive the relative entropy between the probability density functions associated with the two test hypotheses under low-resolution ADCs. We formulate our optimization problem by maximizing the relative entropy, subject to the constant envelope and orthogonality constraints. To suboptimally solve the resultant problem, a two-stage framework is developed. In the first stage, we optimize the transmit power at the directions of the target and clutter. In the second stage, an efficient iterative algorithm based on majorization-minimization is presented to obtain the constant-envelope beamformer according to the attained transmit power. Specifically, we apply a quadratic function as the minorizer, leading to a low-complexity solution at each iteration. In addition, to further facilitate low-cost implementation of the constant-envelope beamformer, we consider the problem of one-bit beamforming design and propose an efficient iterative method based on the Nesterov-like gradient method to solve it. Numerical simulations are provided to demonstrate the effectiveness of the proposed schemes.

Gradient modulated contrastive distillation of low-rank multi-modal knowledge for disease diagnosis.

The fusion of multi-modal data, e.g., medical images and genomic profiles, can provide complementary information and further benefit disease diagnosis. However, multi-modal disease diagnosis confronts two challenges: (1) how to produce discriminative multi-modal representations by exploiting complementary information while avoiding noisy features from different modalities. (2) how to obtain an accurate diagnosis when only a single modality is available in real clinical scenarios. To tackle these two issues, we present a two-stage disease diagnostic framework. In the first multi-modal learning stage, we propose a novel Momentum-enriched Multi-Modal Low-Rank (M3LR) constraint to explore the high-order correlations and complementary information among different modalities, thus yielding more accurate multi-modal diagnosis. In the second stage, the privileged knowledge of the multi-modal teacher is transferred to the unimodal student via our proposed Discrepancy Supervised Contrastive Distillation (DSCD) and Gradient-guided Knowledge Modulation (GKM) modules, which benefit the unimodal-based diagnosis. We have validated our approach on two tasks: (i) glioma grading based on pathology slides and genomic data, and (ii) skin lesion classification based on dermoscopy and clinical images. Experimental results on both tasks demonstrate that our proposed method consistently outperforms existing approaches in both multi-modal and unimodal diagnoses.

ContextLoc++: A Unified Context Model for Temporal Action Localization.

Effectively tackling the problem of temporal action localization (TAL) necessitates a visual representation that jointly pursues two confounding goals, i.e., fine-grained discrimination for temporal localization and sufficient visual invariance for action classification. We address this challenge by enriching the local, global and multi-scale contexts in the popular two-stage temporal localization framework. Our proposed model, dubbed ContextLoc++, can be divided into three sub-networks: L-Net, G-Net, and M-Net. L-Net enriches the local context via fine-grained modeling of snippet-level features, which is formulated as a query-and-retrieval process. Furthermore, the spatial and temporal snippet-level features, functioning as keys and values, are fused by temporal gating. G-Net enriches the global context via higher-level modeling of the video-level representation. In addition, we introduce a novel context adaptation module to adapt the global context to different proposals. M-Net further fuses the local and global contexts with multi-scale proposal features. Specially, proposal-level features from multi-scale video snippets can focus on different action characteristics. Short-term snippets with fewer frames pay attention to the action details while long-term snippets with more frames focus on the action variations. Experiments on the THUMOS14 and ActivityNet v1.3 datasets validate the efficacy of our method against existing state-of-the-art TAL algorithms.

Effect of Spatial and Spectral Scaling on Joint Characterization of the Spectral Mixture Residual: Comparative Analysis of AVIRIS and WorldView-3 SWIR for Geologic Mapping in Anza-Borrego Desert State Park.

A geologic map is both a visual depiction of the lithologies and structures occurring at the Earth's surface and a representation of a conceptual model for the geologic history in a region. The work needed to capture such multifaced information in an accurate geologic map is time consuming. Remote sensing can complement traditional primary field observations, geochemistry, chronometry, and subsurface geophysical data in providing useful information to assist with the geologic mapping process. Two novel sources of remote sensing data are particularly relevant for geologic mapping applications: decameter-resolution imaging spectroscopy (spectroscopic imaging) and meter-resolution multispectral shortwave infrared (SWIR) imaging. Decameter spectroscopic imagery can capture important mineral absorptions but is frequently unable to spatially resolve important geologic features. Meter-resolution multispectral SWIR images are better able to resolve fine spatial features but offer reduced spectral information. Such disparate but complementary datasets can be challenging to integrate into the geologic mapping process. Here, we conduct a comparative analysis of spatial and spectral scaling for two such datasets: one Airborne Visible/Infrared Imaging Spectrometer-Classic (AVIRIS-classic) flightline, and one WorldView-3 (WV3) scene, for a geologically complex landscape in Anza-Borrego Desert State Park, California. To do so, we use a two-stage framework that synthesizes recent advances in the spectral mixture residual and joint characterization. The mixture residual uses the wavelength-explicit misfit of a linear spectral mixture model to capture low variance spectral signals. Joint characterization utilizes nonlinear dimensionality reduction (manifold learning) to visualize spectral feature space topology and identify clusters of statistically similar spectra. For this study area, the spectral mixture residual clearly reveals greater spectral dimensionality in AVIRIS than WorldView (99% of variance in 39 versus 5 residual dimensions). Additionally, joint characterization shows more complex spectral feature space topology for AVIRIS than WorldView, revealing information useful to the geologic mapping process in the form of mineralogical variability both within and among mapped geologic units. These results illustrate the potential of recent and planned imaging spectroscopy missions to complement high-resolution multispectral imagery-along with field and lab observations-in planning, collecting, and interpreting the results from geologic field work.

A two-stage feature aggregation network for multi-category soybean leaf disease identification

Accurate identification of soybean leaf disease is of utmost importance for its cultivation and fine management, as it is a critical factor contributing to the decreased quality and yield of soybean. Nevertheless, several existing studies on the identification of soybean leaf disease encounter the challenge of striking a balance between model expressiveness and practical applicability. To address such issue, we propose a well-designed two-stage feature aggregation network framework (TFANet). The main component of it is the two-stage feature aggregation (TFA) module, which is designed to achieve great feature representation capability by aggregating feature information from different convolutional layers in two stages. Meanwhile, TFA module incorporates the efficient channel attention to extract crucial disease information. Afterwards, to limit the loss of feature information and expand the receptive field, a dilated convolution-based feature fusion (DCFF) module is developed. Additionally, the InceptionC module is introduced to further achieve better performance. The experimental results demonstrated that TFANet obtained an accuracy of 98.18% and an F1 scores of 98.39%, with merely 1.18 M parameters. In comparison with some classic convolutional neural network models, TFANet showed remarkable superiority in multiple evaluation metrics. Hence, TFANet has great potential to be used in practical soybean leaf disease identification.

A two-stage unsupervised approach for surface anomaly detection in wire and arc additive manufacturing

Wire and arc additive manufacturing (WAAM) has gradually been applied in industrial applications in recent years due to its low cost, high deposition rate, and high material utilization rate. Anomalies in the WAAM process, such as inclusion, porosity, and lack of fusion, can have unpredictable effects on the quality of the final product. While some studies have investigated anomaly detection methods in the WAAM process, they mainly rely on supervised learning methods that require extensive manual labeling, with less attention paid to unsupervised models. Furthermore, most studies focus on significant anomalies that are rare in actual production, limiting their practical application. This paper proposes a two-stage unsupervised defect detection framework based on online melt pool video data. By considering the motion characteristics of the manufacturing process, a revised threshold method is used to detect anomalies during the WAAM process. Combining machine contextual information, the physical spatial location of defects is further identified and displayed through a human-machine interactive interface. The dataset used in this study is derived from real printing processes of WAAM parts. Compared with baseline methods, the proposed approach significantly improves recall and achieves an F1-score of 86.3% on the test set.

A Two-stage Diagnostic Framework for Post-ablation Treatment Response Assessment in Patients with Hepatocellular Carcinoma.

Contrast-enhanced ultrasound (CEUS) video plays an important role in post-ablation treatment response assessment in patients with hepatocellular carcinoma (HCC). However, the assessment of treatment response using CEUS video is challenging due to issues such as high inter-frame data repeatability, small ablation area and poor imaging quality of CEUS video. To address these issues, we propose a two-stage diagnostic framework for post-ablation treatment response assessment in patients with HCC using CEUS video. The first stage is a location stage, which is used to locate the ablation area. At this stage, we propose a Yolov5-SFT to improve the location results of the ablation area and a similarity comparison module (SCM) to reduce data repeatability. The second stage is an assessment stage, which is used for the evaluation of postoperative efficacy. At this stage, we design an EfficientNet-SK to improve assessment accuracy. The Experimental results on the self-collected data show that the proposed framework outperforms other selected algorithms, and can effectively assist doctors in the assessment of post-ablation treatment response.