2 research outputs found
NUMERICAL ANALYSIS OF IMMERSION COOLING ON A SERVER USING AL2O3/MINERAL OIL NANOFLUID
No full textIn the current world, technology has developed significantly, with a massive processing and storage of data, resulting in the high increase in Power Density and Heat Generation of servers, computers and its components in data centers. This calls for an engineering solution, for efficient heat dissipation of these servers to ensure their reliability and prolonged working. Air cooling is a prominent method of data center cooling. However, due to its low heat carrying capacity, it is not an efficient method for cooling high heat generating servers. There are two methods to remove this heat, increasing the area of heat transfer being one of the methods, which is not feasible everywhere. To tackle this problem, liquid immersion cooling method has emerged as a prominent method for cooling servers and its components in data centers, where the servers can directly be immersed inside the liquid, making the process simpler and cost effective. Water has higher thermal properties like heat capacity, but the limitation is that the liquid must be dielectric to save equipment from short-circuit. This feature also influences the thermal conductivity of the liquids. Generally dielectric liquids have low thermal conductivity which affects the thermal performance of the cooling process. Thermal conductivity of the dielectric liquids is drastically increased with the introduction of nano particles, which has proven to be the best method. Nano particles are metal and non-metal particles with the size between 1 to 150 nano-meters. To keep the dielectric feature of the liquid, the non-metallic nano particle can be added to the liquid. Therefore, the Alumina which is one of the materials using as an electrical insulation is used. The mean size of the nano particle used in this work is 80 nm and the liquid are mineral oil, which is a low-price liquid for immersion cooling compared to other dielectric liquids. The concentration of nano particles was 0 to 5 percent and it is assumed to remain homogeneous. The properties of the mixture were calculated based on the theoretical formula and it was function of temperature. In this work, we simulated heat transfer and effect of the nano particle concentration on the junction temperature of the processors using CFD techniques. The chosen server is an open compute server which has two processor in a row. The server was modeled in Ansys Icepack and simulations were performed for pure mineral oil and nano-fluid at particle concentrations of 1%, 3% and 5 % at 1, 2 and 3 LPM respectively. Prior to the study, effect of frame height on the maximum CPU junction temperature was tested using pure mineral oil. Drastic reduction in maximum CPU temperature was observed with a smaller frame height and this height was maintained throughout the study. Simulations were conducted for 3 different heatsink geometries, namely parallel plate, cylindrical bonded pin and plate fin heatsink. From results obtained, effect of nano-particle concentration on the maximum CPU junction temperature, pressure drop and pumping power were studied and comparisons were made for different nano-particle concentration and flow rates. Also, comparisons on CPU junction temperature, pressure drop, and pumping power obtained from simulations using the 3 different heatsink geometries were made
Comprehensive analysis of the impact of pre-processing techniques on the performance of a cnn-based facial age estimation model
No full textFacial age estimation has been a complex task in the field of computer vision due to the factors such as variations in the facial appearance, complex aging pattern and the lack of a standard knowledge as to the basis of what really constitutes towards calculating the age using facial features. With the ever improving technology, Convolutional Neural Networks (CNNs) have been proven to be the most powerful solution the facial age estimation problem. We aslo need to note that the performance of these models heavily depend on the the quality of the images that they are trained on. Fortunately, certain pre-processing techniques in the field have shown that they can improve the performance of these models by enhancing certain features in a photo. In this research paper, we try to provide a comprehensive analysis of the impact of various, commonly used pre-processing techniques in the computer vision field over the performance of a CNN-based facial age estimation model. Specifically, we investigate the effectiveness of Image Rescaling, Histogram Equalization, Gaussian smoothing, and Grayscale conversion techniques on three widely used datasets, namely UTKFace, MegaAge, and FGNet. Our experimental results indicate that preprocessing techniques have a significant impact on the performance of the age estimation model. Among the techniques studied, Histogram Equalization is found to be the most effective in improving the accuracy of the model on all three datasets. and Gaussian smoothing and Image Rescaling techniques come close to the positive impact of Histogram Equalization but just fall short. Image Rescaling was successful only on the FGNet dataset. Although it didn't show a significant improvement in the performance of the model, it improved its time and memory used during the compilation. Furthermore, it was noticed that the impact of these pre-processing techniques varied depending on the dataset being considered. For instance, the Histogram Equalization technique is found to be most effective on UTKFace. Whereas, Image Rescaling came out on top for the FGNet datasets and Gaussian smoothing showed it is most effective on the MegaAge dataset. The findings in our work contain important comparisons for the researchers in the field of computer vision especially in the field of facial age estimation. This work will give them an idea and guidance towards picking the best pre-processing techniques to enhance the quality of the images for the specific dataset they're using enabling them to extract the maximum potential of the the age estimation they're using
