Submissões Recentes
- Carregando...
Low-resolution neural networks
2025 - CABRAL, EDUARDO L.L.; DRIEMEIER, LARISSA
The expanding scale of large neural network models introduces significant challenges, driving efforts to reduce memory usage and enhance computational efficiency. Such measures are crucial to ensure the practical implementation and effective application of these sophisticated models across a wide array of use cases. This study examines the impact of parameter bit precision on model performance compared to standard 32-bit models, with a focus on multiclass object classification in images. The models analyzed include those with fully connected layers, convolutional layers, and transformer blocks, with model weight resolution ranging from 1 bit to 4.08 bits. The findings indicate that models with lower parameter bit precision achieve results comparable to 32-bit models, showing promise for use in memory-constrained devices. While low-resolution models with a small number of parameters require more training epochs to achieve accuracy comparable to 32-bit models, those with a large number of parameters achieve similar performance within the same number of epochs. Additionally, data augmentation can destabilize training in low-resolution models, but including zero as a potential value in the weight parameters helps maintain stability and prevents performance degradation. Overall, 2.32-bit weights offer the optimal balance of memory reduction, performance, and efficiency. However, further research should explore other dataset types and more complex and larger models. These findings suggest a potential new era for optimized neural network models with reduced memory requirements and improved computational efficiency, though advancements in dedicated hardware are necessary to fully realize this potential.
- Carregando...
Thermoluminescence, electron paramagnetic resonance, and structural characterization of natural prehnite for high-dose radiation dosimetry
2026 - GOMES, MONISE B.; GONZALES-LORENZO, CARLOS D.; ROCCA, RENE R.; RAMIREZ, F. N.; CUEVAS-ARIZACA, EDY E.; SILVA-CARRERA, BETZABEL N.; GUNDU RAO, T. K.; CANO, NILO F.; CHUBACI, JOSE F. D.
Natural prehnite was evaluated as a thermoluminescence (TL) material for high-dose dosimetry. XRF identified SiO₂, Al₂O₃, and CaO as major constituents. Prehnite samples annealed between 200 and 800 °C (1 h) were examined by XRD and Rietveld analysis, confirming prehnite as the majority phase and revealing calcite and vaterite after heating. The 600 °C sample provided the highest TL yield and was selected for detailed study. Fading tests showed a ~40% loss in the 240 and 350 °C peaks over 5.42 days (130 h), followed by signal stability. Dose–response demonstrated, for the 245 °C peak and 325 °C peak, linear regions in the ranges of 0.1–2 kGy and 0.5–20 kGy, respectively, with saturation occurring between 30 and 100 kGy. Kinetic parameters were obtained using TM-Tstop, initial-rise, and variable-heating-rate methods; glow-curve deconvolution resolved five TL components. TL spectra displayed an intense band near 530 nm and a weak band at 550 nm, suggesting two main recombination centers.
- Carregando...
Thermodynamic integration of HTGR nuclear heat into the Barrancabermeja refinery
2026 - ENCISO, NICOLAS; MORALES, DANIEL
Decarbonizing refinery process heat and hydrogen production remains a major challenge for industrial net-zero pathways. High-temperature gas-cooled small modular reactors (HTGR–SMRs) are promising candidates because they combine high outlet temperatures, modular deployment, and inherent safety. This work develops a unified thermodynamic–exergy framework based on Pinch Point analysis, differential heat-transfer integration, and entropy-generation balances to assess HTGR coupling with refinery heat, hydrogen, and cogeneration systems under realistic industrial conditions. Results show that the main thermodynamic difference among the evaluated configurations is not the presence of the Once-Through Steam Generator (OTSG) itself, but the thermal matching between the nuclear heat source and refinery demands. Indirect steam delivery through an OTSG and direct helium supply to multiple refinery processes achieve comparable refinery-level performance, with second-law efficiencies of approximately 62.57%. In contrast, direct helium supply to the topping process—one of the most critical units in the refinery—reaches a lower efficiency of about 57.5%, corresponding to a relative improvement of approximately 8.8% for the OTSG and multi-process configurations. This indicates that, although direct helium exchange preserves a higher temperature potential, it suffers from greater thermal mismatch and higher irreversibility at the reactor–process interface, particularly when applied to single high-demand units such as topping. Applied to the Barrancabermeja Refinery, the proposed architecture could supply about 400 MWth of process heat while supporting cogeneration, low-carbon hydrogen pathways, and avoiding on the order of 3 Mt CO₂/year.
- Carregando...
Analysis of the influence of geometric parameters of laser-produced texturing on carbide tools in Ti6Al4V turning
2026 - SOUZA, FELIPE C. R. de; PAULA, FABIO R. de; MACHADO, ALISSON R.; ROSSI, WAGNER de
This study focuses on analyzing the influence of different groove texture geometries on carbide cutting tools for turning Ti6Al4V titanium alloy, exploring how different parameters (depth, width, spacing, direction, and distance from the cutting edge) affect machining performance. This work contributes to the advancement of machining techniques, promoting more efficient and sustainable processes, with potential practical applications in industry. The objectives include reducing machining forces while minimizing machining costs. The methodology employed advanced technologies, such as femtosecond lasers, to produce defect-free textures without thermal damage or microcracks. Initial results demonstrated that textured tools can significantly improve the machining process by reducing machining forces. The results showed that texture direction and distance from the cutting edge were the most influential parameters, while texture dimensions had a lesser impact. Furthermore, the effects of textures were found to vary with feed rate, highlighting the need for specific optimizations for different cutting regimes (such as roughing or finishing).