Constantly reducing the size and improving the performance of electronics lead to the heat generation enhancement and unstable operation of such devices. Therefore, the modeling and production of the efficient cooling systems is one of the most important modern challenges. Based on this, in the present research a numerical study on the heat-conducting base material influence on the power-law nanofluid thermal convection within a cavity having a local source of constant volumetric heat generation is performed. The problem is described by transient differential equations written using the Oberbeck–Boussinesq approximation, the Ostwald-de Waele power law and the single-phase nanoliquid approaches. The nanosuspension characteristics are calculated employing the experimental relations reflecting the impact of the nanoparticles concentration and liquid temperature. The equations are solved using the developed in-house numerical algorithms based on the finite difference technique. The locally one-dimensional difference scheme has been used to reduce 2D problem to a set of 1D problems. Convective terms have been approximated by the second upstream scheme, the diffusion terms are approximated by the central differences, while the transient terms are discretized by the Euler scheme. The obtained sets of algebraic equations have been solved using effective direct and iterative methods. The work analyzes the impact of the power-law index, the material of the thermally conductive substrate, the material of nanoparticles and their concentration on heat transfer performance and the cooling technique efficiency. Numerical simulations show that the simultaneous usage of a thermally conductive copper substrate and copper nanoparticles makes it possible to diminish the mean temperature of the heated element. Moreover, a growth of the nanoparticles volume fraction allows to decrease the heater average temperature. The obtained results can be used for improvement of the electronics cooling systems, modern heat exchangers and thermal energy storage systems where the combination of non-Newtonian nanofluid as a working fluid and heat-conducting substrates can be considered as an effective approach for the optimization of the heat transfer.
Accurate left ventricle segmentation in echocardiography is essential for assessing cardiac function and diagnosing cardiovascular disease. Although recent deep learning methods have shown promising performance, several challenges remain. Conventional U-Net-based architectures are effective at learning local spatial representations but have limited ability to model long-range dependencies in noisy ultrasound images. In contrast, Transformer-based methods capture global context but often sacrifice fine boundary details and computational efficiency. The shape, size, and echogenicity of ventricular structures vary considerably across patients, while myocardial borders are frequently degraded by acoustic shadowing and speckle noise, making precise delineation difficult. To address these limitations, we propose W-KANet, a novel segmentation framework composed of three specialized modules. First, Residual Squeeze-Dilated Blocks (RSDBlocks) employ cascaded dilated convolutions with channel attention for hierarchical feature aggregation. Second, Wavelet-guided Kolmogorov–Arnold Networks with Grouped Multi-Scale Attention Blocks (WKGMSABlocks) utilize Haar wavelet decomposition, where low-frequency subbands guide learnable base activations for semantic representation, while high-frequency subbands guide B-spline activations for boundary enhancement, with grouped attention that scales linearly with the number of tokens. Third, Tri-Stream Boundary-Reverse Fusion (TSBRF) integrates encoder features through dual ASPP fusion, reverse attention for background suppression, and multi-scale boundary detection. Experiments on five datasets (CAMUS, EchoNet-Dynamic, MCE, HMC-QU, ACDC) demonstrate competitive performance, achieving 92.53%, 93.91%, 86.47%, 94.91%, and 92.71% Dice scores respectively, outperforming UNet, TransUNet, U-Mamba, and U-KAN across evaluation metrics. Our code is available at: https://github.com/Hassan48khan/W-KANet.
Approximate nearest-neighbor (ANN) indexes accelerate candidate retrieval, but comparisons at unrelated retrieval fidelity can produce misleading system choices. We evaluate exact Flat, HNSW, IVF-Flat, IVF-PQ, and Flat-PQ on four public interaction datasets and separate user-to-item from item-to-item retrieval. The CPU-only protocol calibrates tunable indexes against exact Flat at agreement-recall targets of 0.90, 0.95, and 0.98. A held-out-item rank proxy is reported separately from exact-neighbor fidelity. Because the archived pipeline learns item representations from the complete filtered interaction matrix before constructing leave-one-out queries, this proxy is transductive and is not an estimate of out-of-sample recommender quality. The study combines 40 primary conditions with query-level discordance, corrected paired inference over three nonredundant outcomes, exposure proxies, SVD-weighting sensitivity, and storage-and-fidelity synthetic scaling. HNSW and IVF-Flat reached all targets on the calibration samples, whereas the tested IVF-PQ configuration reached 0.90 in two of eight scenarios and never reached 0.95. Expanded-sample evaluation at the selected 0.95 setting retained the target in seven scenarios for HNSW and six for IVF-Flat, revealing finite-sample transfer uncertainty. HNSW and IVF-Flat changed the held-out item’s top-10 rank for only 0.03% and 0.11% of queries on average, compared with 5.23% for IVF-PQ and 6.01% for Flat-PQ. The corrected 20,000-resample analysis rejected 18 of 96 Holm-adjusted comparisons, predominantly for Recall@100; every rejection favored exact Flat. The results support constraint-first index selection in which the agreement threshold is derived from tolerated degradation in an application-specific downstream or online service metric, not treated as a universal technical constant.
This paper presents a detailed experimental investigation of the near-wall transport dynamics in Taylor-Couette-Poiseuille flows using measurements of the large-scale variations of local wall-shear-rate components. A three-segment electrodiffusion probe was used to conduct multiple experiments across a wide range of Taylor numbers (300 < Ta <2520) and Reynolds numbers (2 < Re <32) to capture laminar, transitional, and turbulent flow conditions. Both azimuthal and axial wall shear rate component measurements were made in order to directly correlate the effects of flow structures on the transport of momentum in the near-wall region. The results show that the average azimuthal wall shear rate is primarily influenced by the Taylor number, while its variation with the Reynolds number remains within the experimental uncertainty. On the other hand, the large-scale shear rate dynamics are strongly influenced by the flow structures. More specifically, vortical flows have almost identical average wall shear rates as helical flows but distinctly different temporal characteristics and fluctuation magnitude. The differences in large-scale wall shear dynamics between the two flow structures arise from distinct mechanisms of momentum transfer; vortices induce high shear rates through rapid radial inflow and outflow, whereas helices generate low shear rates via a much more uniform shear distribution due to their oblique configuration. These findings suggest that the average wall shear stress alone is insufficient to determine the transport processes occurring in the wall region. Localized measurements are required in order to establish the relationship between flow structure and transport dynamics within the measurement bandwidth. These findings provide new insight into shear-induced transport in rotating flows and establish experimental benchmarks for the development and validation of transport models.
The increasing demand for durable and high-performance pavements has led to the widespread use of polymer-modified asphalt (PMA) mixtures owing to their enhanced elasticity, rutting resistance, and fatigue performance. However, the complex multi-phase and heterogeneous nature of these materials limits the effectiveness of conventional experimental and continuum-based modeling approaches in fully capturing their mechanical behavior. The discrete element method (DEM) has emerged as an effective mesoscale numerical tool for modeling particle interactions, contact behavior, and internal structural development in asphalt mixtures. This paper presents a critical state-of-the-art review of DEM-based modeling of polymer-modified asphalt mixtures, focusing on material properties, modeling techniques, and applications. It examines polymer modification mechanisms, including network formation, viscoelastic behavior, temperature susceptibility, and aging effects, and assesses their implications for DEM contact modeling and parameter calibration. Because the literature identified for this review is overwhelmingly dominated by styrene-butadiene-styrene (SBS) modification, findings are explicitly flagged throughout as SBS-specific or as more broadly applicable to PMA. In this review, DEM model-generation techniques, contact constitutive laws, and calibration strategies are compared, and key applications, including compaction, rutting, fatigue cracking, moisture damage, and pavement maintenance, are discussed. A bibliometric analysis was used to map current research trends and identify gaps, particularly the limited incorporation of polymer microstructures and rheology into DEM models. This review outlines priority directions for future research, including multiscale modeling, DEM-FEM coupling, CT-based reconstruction, and AI-assisted parameter calibration, and proposes a framework linking polymer-modified material behavior to DEM modeling requirements to support future improvements in simulation accuracy and pavement design.
