parallel processing

In the present study, the new ADI-CEIDD approach is proposed by combining the ADI method with the explicit-implicit domain decomposition method. The method is used for solving the two-dimensional conduction heat transfer equation on GPU .In this method, an explicit numerical scheme is used to predict values at inner boundaries and an implicit scheme based on the ADI method is used to solve the sub-domains. Then, an implicit scheme is used to correct the values on the inner boundary. The present method increases the number of independent sets of equations and enables more threads to occupy the device. Numerical experiments are done to investigate the accuracy and speed of the method. The results show that the ADI-CEIDD can achieve a speedup of 1.3 to 2.6 times compared to the ADI method. By increasing the number of subdomains from 2 to 32, the speed of the proposed method is increased up to 1.6 times and the accuracy decreases. Although the error of the presented method is higher than the ADI method, numerical experiments show high stability of the ADI-CEIDD. Furthermore, the results show that the ADI-CEIDD method is more advantageous to problems with coarse grid. By increasing the grid size from 256 * 256 to 512 * 512, the value of the Sp decreases from 2.4 to 1.7.

Analysis of natural gas transient flow in transmission pipelines is one of the most important issues in the gas industry. Despite the previous studies, the accuracy and the computational time have yet considered as two important challenges in this field. In this paper, a parallel algorithm for numerical simulation of isothermal and non-isothermal gas flows is presented. Numerical analysis of the flow is performed using the implicit Steger-Warming flux vector splitting method. For parallelization, the computer program has been parallelized using Message Passing Interface library. In order to demonstrate the capabilities of the developed computer program, the flow inside two pipelines with different conditions is solved, and the results are validated. Then, some factors such as the computational time, reduction of the time, and the speed up criteria are obtained to demonstrate the computational efficiency of the proposed method. The results show that parallel processing method can significantly reduce computational time of natural gas flow in long transmission pipelines. Moreover, it is shown that application of this approach on the fine computational grids is more efficient than on the coarse grids.

The main goal of the present paper is development of the unsteady turbulence modeling using the URANS algorithm and preservation of numerical performance and assessment of this method respect to the RANS model in numerical simulation of a sonic jet in supersonic cross flow. The turbulence modeling used in both algorithms is the Spalart Almaras model. To improve accuracy of the computations, the structured multi block grid is used and to decrease the computational cost, the OMP parallel processing is applied. In this paper, firstly, the governing equations of both the RANS and URANS are described and then the developed code is used to analyze a 3D jet in cross flow. The results including flow structure, distribution of the pressure and velocity profile are compared with experimental data. The URANS method show more accurate results than the RANS model in numerical simulation of the sonic jet in supersonic cross flow.

In this paper three algorithms of Cyclic-Reduction, Parallel-Cyclic-Reduction and Parallel-Thomas are introduced to solve the Tridiagonal system of equations using GPUs and the effect of coalesced-memory-access and uncoalesced-memory-access to global memory are studied. To assess the ability of these algorithms, as a case-study the simulation of the lid-driven cavity flow have been compared to the results of Runtimes and physical parameters of the classical Thomas algorithm, executed on CPU. The maximum speed-up of these algorithms against CPU runtime is about 4.4x, 5.2x and 38.5x, respectively. Also, approximately a 2x speed-up achieved in coalesced-memory access on GPU.

Investigation of fluid-solid interaction has been studied as an introduction to simulate a wide range of engineering problems such as fluidized beds, sediment transportation and catalyst inks in fuel cells. An efficient method for performing such simulations is a combination of Lattice Boltzmann method (LBM) and Smoothed Profile Method (SPM). In addition, the operations in the SPM are local; it can be easily programmed for parallel processing. In this approach, the flow is computed on fixed Eulerian grids which are also used for the particles. Owing to the use of the same grids for simulation of fluid flow and particles, this method is highly efficient for purpose of parallel processing by means of GPU. In this study, a combination of Lattice Boltzmann method (LBM) and Smoothed Profile method has been implemented in parallel processing on GPU. For validation purpose, the fluid flow within a channel was investigated. Results suggest that computational time can be reduced up to 80 times by means of GPU.Then, drag force exerted on a sphere in fluid flow and the sedimentation of one sphere in a quiescent fluid were studied. Results show that performance of GPU can be increased up to 6.5 million fluid nods per second by using this method.

Predicting wind flow pattern around high rise building, because of pedestrian comfort, air pollution in weak wind region and etc. has important position in wind engineering. Turbulent wind flow over buildings due to the complexity like sharp corners, ground effect and different vortexes, is one of the best choices to evaluate turbulence methods. Moreover in a campus due to high velocity region between buildings, simulating wind flow is more complex. Therefore reaching acceptable result needs a fine grid with an accurate turbulence model that increases computational cost. DES is hybrid RANS-LES models for simulating turbulent flow which for their characteristic, treat near wall as RANS and farther the wall act as LES model. Consequently in this hybrid model, computational time will decrease compared to traditional LES models. In this article turbulent 3 dimensional wind flow over Tarbiat Modares University with DES method in different wind velocities is simulated. Because cells number is great, parallel processing has been used. For verification, DES results are compared with traditional LES models such as smagorinsky. The results show good agreement with other traditional methods.

Turbulent wind flow over buildings occurs due to the complexity like sharp corners، ground effect and different vortexes is one of the best choices to evaluate turbulence methods. DES and DDES are hybrid RANS-LES models for simulating turbulent flow which for their characteristic treat near wall as RANS and farther the wall act as LES model. Consequently computational time will decrease compared to traditional LES models. In this article to evaluate DES and DDES models، turbulent incompressible flow in Re = 22000 over 3D building is simulated using parallel processing facilities. For verification purpose other investigators experiment results are used. Also the mentioned models are compared with classic RANS and LES models، like k-ε and LES-Smagorinsky to depict their performance. Our results illustrate DES model with fine grid has good precision for simulating turbulent incompressible wind flow over building and decline of 26 percentage of computational time compared to LES-Smagorinsky model.