Analyses are carried out to study the effects of air velocity, particle circulating rate and fuel/air ratio on the profiles of temperature and heat flux in circulating fluidized bed boilers. The profile of particle concentration within the furnace is obtained by assuming the flow of gas and particles in the system as one-dimensional solid-gas two-phase flow and by considering that the particles are fed from the underside of the furnace with zero velocity, and are accelerated by the up-ward air flow. The radiative heat transfer in the two-phase flow system is solved by using the Monte Carlo method. In the heat-transfer analysis, the walls surrounding the system are considered as gray walls, and the two-dimensional heat transfer analysis is carried out in the one-dimensional two-phase flow field.
Scientific communities are increasingly adopting machine learning and deep learning models in their applications to accelerate scientific insights. High performance computing systems are pushing the frontiers of performance with a rich diversity of hardware resources and massive scale-out capabilities. There is a critical need to understand fair and effective benchmarking of machine learning applications that are representative of real-world scientific use cases. MLPerf is a community-driven standard to benchmark machine learning workloads, focusing on end-to-end performance metrics. In this paper, we introduce MLPerf HPC, a benchmark suite of large-scale scientific machine learning training applications driven by the MLCommons Association. We present the results from the first submission round, including a diverse set of some of the world's largest HPC systems. We develop a systematic framework for their joint analysis and compare them in terms of data staging, algorithmic convergence, and compute performance. As a result, we gain a quantitative understanding of optimizations on different subsystems such as staging and on-node loading of data, compute-unit utilization, and communication scheduling, enabling overall $>10 \times$ (end-to-end) performance improvements through system scaling. Notably, our analysis shows a scale-dependent interplay between the dataset size, a system's memory hierarchy, and training convergence that underlines the importance of near-compute storage. To overcome the data-parallel scalability challenge at large batch sizes, we discuss specific learning techniques and hybrid data-and-model parallelism that are effective on large systems. We conclude by characterizing each benchmark with respect to low-level memory, I/O, and network behavior to parameterize extended roofline performance models in future rounds.
The aorta is the thickest blood vessel in the human body. It is a strong and elastic blood vessel consisting of three layers: the intima, the tunica media, and the adventitia. Aortic dissection occurs when the intima tears and a new channel (false lumen) is created. Dissections may lead to various pathological conditions such as aneurysms and rupture. Recently, blood flow analysis has been looked into as a diagnostic tool to determine early stages of aortic dissection. Therefore, a Fluid-Structure Interaction (FSI) analysis between the blood vessel and blood flow of the aorta was conducted to investigate its applicability to a dissection.
Ice slurry has been widely utilized as thermal storage medium due to its high storage density of thermal energy and liquidity. In the heat exchange process of ice slurry through the heated exchanger, its high storage density would not be fully used because the ice particle may not contact to the heated surface so much. In the present study, horizontal heated tubes with longitudinal fins have been adopted as one of the method to enhance the melting of ice particles in the ice slurry. They were arranged in a lattice grid and attack angles of fins were set to control the ice particle flow. A Moving Particle Simulation method (MPS method) has been employed to simulate the flow and heat transfer characteristics of ice particles. The effect of fin array on the energy efficiency of the heat exchanger which is evaluated from both heat transfer enhancement and pressure loss was investigated. As a result of investigation, it was found that horizontal heated tube with longitudinal fins increase its energy efficiency depending on the attack angle array, while they make the contact between ice particle and heated surface decrease comparing to the case of horizontal heated tubes without fins.
As smart phones become more complex, higher in performance and smaller in size, heat concentration at localized areas is becoming a problem. Phase change materials (PCM) have drawn attention as passive thermal management. In this study, simple experiments using PCM were conducted. The results showed that the saturation time increased with increase in the mass of PCM, and a relationship between the saturation time and the mass of PCM was a linear relation. Furthermore, the saturation time increased with decrease in the melting temperature of PCM. As a result, PCM is effective for the thermal management of smart phones.
Combined radiative-convective heat transfer is analyzed in a two-dimensional coaxial jet of nongray gas. The Monte Carlo method is used for the radiative heat transfer analysis. In addition to the nongray analysis, the analysis based on the conventional gray gas assumption is also carried out. From these studies, the gray analysis is found to give higher jet temperature at the nozzle-outlet region and lower temperature at the down-flow region compared with the nongray analysis. The effects of the temperature, velocity, absorbing gas concentration of efflux gas, and the nozzle diameter are also studied using the profile of the jet temperature and of the radiative heat flux emitted from the jet.
Numerical simulation of convection heat transfer from airfoil-shaped tube by using OpenFOAM. OpenFOAM is a C++ toolbox for the development of customized numerical solvers and processing utilities for the solution of heat transfer. Flow and temperature fields are calculated and compared with those of cylindrical heat tube. Reynolds number based on equivalent diameter and approaching velocity is set to be 10, 100, 200. So the field is assumed to be two- dimensional flow. It is found that the ratio of Stanton number and drag coefficient St/CD of airfoil-shaped tube is larger than that of cylindrical tube.
By using the Monte Carlo method, radiative energy transfer through fibrous layer is analyzed. For the material of the layer, glass and nickel fibers are assumed. Mie scattering is assumed for the scattering mechanism by the fibers. The validity of this method is shown by comparing the results with the experiment obtained by Tong et al. The heat fluxes through the layer for various temperature differences over the layer and for different thicknesses are predicted well by the method. In the present study, the directional profiles of the scattered radiative energy are obtained to study the effect of multiple-scattering. Two types of fiber orientations are considered: a three-dimensionally randomly oriented layer, and a layer which is oriented randomly in two-dimension and is parallel to the bounding surface. From the analysis, following results are obtained. (1) The heat flux through the layer with three-dimensionally randomly oriented fibrous layer is about 1.3 times larger than that with two-dimensionally randomly oriented layer. (2) The directional profile of scattered energy within the layer changes when the number of scattering is increased. (3) The average extinction efficiency for the radiative energy which has the above mentioned directional profile is almost the same for the layersmore » with the two types of orientations. (4) The difference in the directional profile of scattered energy is the main cause to give the difference in the heat flux passes through the layer with different fiber orientations.« less
