Abstract This study investigates the flow characteristics and thermal performance of Straight Vortex Tubes (SVTs) and Convergent–Divergent Vortex Tubes (CDVTs) under identical operating conditions using computational fluid dynamics (CFD). A three-dimensional model with compressed air as the working fluid was simulated using the finite volume method and the RNG k–ε turbulence model. Grid independence was verified to ensure numerical reliability. Results show that while the straight tube (0°/0°) configuration achieves the highest instantaneous temperature separation (ΔT ≈ 48 K), its performance is highly sensitive to geometric variations and deteriorates with changes in diameter and length. In contrast, the CDVT with fixed 10° convergent and 6° divergent angles demonstrates superior cold outlet temperature reduction and vortex stability for shorter tube lengths (90–130 mm), where compactness and robustness are critical. These findings highlight that geometric modifications do not universally maximize ΔT but provide enhanced stability and efficiency in constrained geometries, offering valuable insights for designing compact, energy-efficient vortex-based cooling systems.
Multimodal generative AI systems like Stable Diffusion, DALL-E, and MidJourney have fundamentally changed how synthetic images are created. These tools drive innovation but also enable the spread of misleading content, false information, and manipulated media. As generated images become harder to distinguish from photographs, detecting them has become an urgent priority. To combat this challenge, we release MS COCOAI, a novel dataset for AI generated image detection consisting of 96000 real and synthetic datapoints, built using the MS COCO dataset. To generate synthetic images, we use five generators: Stable Diffusion 3, Stable Diffusion 2.1, SDXL, DALL-E 3, and MidJourney v6. Based on the dataset, we propose two tasks: (1) classifying images as real or generated, and (2) identifying which model produced a given synthetic image. The dataset is available at https://huggingface.co/datasets/Rajarshi-Roy-research/Defactify_Image_Dataset.
This paper introduces the Dynamic Feature Attribution Framework (DFAF), a rigorous methodology for analysing the temporal evolution of feature importance in neural networks across training epochs and distributional shifts. Unlike conventional explainability methods that provide static post-hoc explanations, DFAF models attribution as a dynamic temporal process, revealing when and how models learn feature representations. Through comprehensive experiments on the UCI Adult Income dataset (48842 samples, 12 features, 50 training epochs) across two architectures (MLP and Transformer), we establish three empirically verified findings: (1) Feature importance undergoes a measurable plasticity–stability phase transition: Plasticity Index = 0.0225 in early training vs. Stability Index = 74806.6 in late training, a 1685× ratio. (2) Attribution is more robust to adversarial perturbation than classification accuracy: at ε=0.10 (FGSM), AARS = 0.889 while accuracy falls to 78.2%. (3) Transformers exhibit 39% higher plasticity than MLPs but converge to comparable stability, showing that architecture governs the path of learning, not its destination. The framework introduces four novel metrics—Plasticity Index, Stability Index, Attribution Drift Score, and the new Adversarial Attribution Robustness Score (AARS)—and provides validated deployment thresholds: confidence threshold 0.049 and convergence threshold 2× median drift. Ablation confirms Integrated Gradients achieves Spearman ρ=0.758 vs. permutation importance, outperforming simple gradient methods (ρ=0.491) by 54%.
There are limited effective ways of monitoring self-directed study. Observation, reports, and questionnaires are subject to recollection bias whereas sensor-based methods like EEG and eye tracking, although accurate, are quite costly, invasive and inaccessible. To address these issues, we suggest a lightweight, web-based application that will use the real-time facial emotion recognition (FER). It combines Blazeface to identify faces, Emotion classifier using FER.js and interactive analytics using Chart.js along with a personalized feedback recommendation engine. The application tested on commodity laptops in various locations delivered high response rates (87 - 92%) and significant focus scores and privacy because it operates as fully client-side.
The Significant Wave Height (SWH) is an essential factor in maritime navigation, port operations and safety of the coastal infrastructure. The correct forecasting of SWH is critical to the reduction of risks of the extreme state of the ocean and ability to control the port environment in a sustainable manner. This paper examines how wind elements (eastwest and northsouth), the speed of wind, and the Sea Surface Temperature (SST) affect SWH in three major Indian ports namely Cochin, Visakhapatnam, and Jawaharlal Nehru Port Trust (JNPT). The data analyzed are based on long-term reanalysis (1979-2009) that is 6 hourly (00, 06, 12, 18 UTC). The relationships between atmospheric and oceanic variables are analyzed using statistical means, correlation, trend, time-series analysis and so on. This is then predicted in a Random Forest machine learning model that effectively excludes linear interactions and regional variability. The findings show that wind speed, wind direction, and SST have strong effects on SWH with a clear difference between the ports. Moreover, an interactive analysis can be obtained with the help of a user-friendly interface created with the help of Streamlit. On the whole, the proposed study combines statistical and machine learning methods to make the prediction of waves more accurate, facilitate the preparedness to coastal hazards, and ensure the safety and efficiency of ports functioning as part of the sustainable coastal management.
