• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.11a
• /
• pp.24-26
• /
• 2022

시스톨릭 어레이는 DNN training 등 인공지능 연산의 대부분을 차지하는 행렬 곱셈을 수행하기 위한 하드웨어 구조로 많이 사용되지만, sparsity 가 높은 행렬을 연산할 때 불필요한 동작으로 인해 효율성이 크게 떨어진다. 본 논문에서 제안된 유동적 시스톨릭 어레이는 matrix condensing, weight switching, 그리고 direct output path 의 방법과 구조를 통해 sparsity 가 높은 행렬 곱셈의 수행 사이클을 줄일 수 있다. 시뮬레이션을 통해 기존 시스톨릭 어레이와 유동적 시스톨릭 어레이의 성능을 비교하였으며 8×8, 16×16, 32×32 의 크기를 가진 행렬을 동일 크기의 시스톨릭 어레이로 연산하였을 때 필요 사이클 수를 최대 12 사이클 절감할 수 있는 것을 확인하였다.

• Tunnel and Underground Space
• /
• v.10 no.2
• /
• pp.173-179
• /
• 2000

The influence of the plastic flow rules on the elasto-plastic behaviour of a discrete joint element was investigated by performing the numerical direct shear tests under both constant normal displacement and normal displacement conditions. The finite interface elements obeying Plesha’s joint constitutive law was used to allow the relative motion of the rock blocks on the joint surface. Realistic results were obtained in the tests adopting the non-associated flow rule, while the associated flow rule overestimated the joint dilation. To overcome the computational drawbacks coming from the non-symmetric element stiffness matrix in the conventional non-associated plasticity, the symmetric formulation of the tangential stiffness matrix for a non-associated joint element was proposed. The symmetric elasto-plastic matrix it derived by assuming an imaginary equivalent joint with associated flow rule which shows the same plastic response as that of original Joint with non-associated flow rule. The validity of the formulation was confirmed through the numerical direct shear tests under constant normal stress condition.

• Transactions of the Korean Society of Mechanical Engineers
• /
• v.16 no.11
• /
• pp.2033-2048
• /
• 1992

This paper presents the theoretical development and qualitative evaluation of a new concept in the mathematical modeling of dynamic structures. We use both test data and analytical approximations to identify the parameters of an incomplete model. The model has the capability of predicting the response of the points of interest on the structure over the frequency range of interest and can be used to predict the changes in natural frequencies and normal modes due to structural changes. The theory was tested by running simulated tests on a relatively simple structure, identifying the parameters of the incomplete model, and using this model to predict the effects on frequency and mode shapes of several mass and stiffness changes. The conditions of the tests were varied by selecting different numbers of points of measurement, varying the frequency range, and by including assumed measurement error. It is recommended that the theoretical development be continued and that applications to more complex structures be carried out in order to develop a better understanding of the limitations and capabilities of the method. A successful, more definitive sevaluation could lead to immediate practical applications.

• The Journal of the Acoustical Society of Korea
• /
• v.16 no.6
• /
• pp.68-75
• /
• 1997

In this study, to investigate vibration response of piping systems due to pulsation flow, a transfer matrix method is presented. Fluid-pipe interaction is formulated using wave equation for flow velocity and pressure, which depends on position and time. From the wave equation, transfer matrix is obtained. The dynamic responses of piping systems induced by pulsation flow appeared to depend upon fluctuation fluid velocity and pressure occurrnece from pulsation, and beating phenomena were observed near the resonance. Consequently, the dynamic behaviors of piping systems appeared to the same as response characteristics of the inside flow pattern of the pipe, and are determined by the inside fluid flow.

• Transactions of the Korean Society of Mechanical Engineers B
• /
• v.40 no.9
• /
• pp.597-604
• /
• 2016

A parallel algorithm of bi-conjugate gradient method was developed based on CUDA for parallel computation of the incompressible Navier-Stokes equations. The governing equations were discretized using splitting P2P1 finite element method. Asymmetric stenotic flow problem was solved to validate the proposed algorithm, and then the parallel performance of the GPU was examined by measuring the elapsed times. Further, the GPU performance for sparse matrix-vector multiplication was also investigated with a matrix of fluid-structure interaction problem. A kernel was generated to simultaneously compute the inner product of each row of sparse matrix and a vector. In addition, the kernel was optimized to improve the performance by using both parallel reduction and memory coalescing. In the kernel construction, the effect of warp on the parallel performance of the present CUDA was also examined. The present GPU computation was more than 7 times faster than the single CPU by double precision.

• Journal of the Korean GEO-environmental Society
• /
• v.15 no.3
• /
• pp.41-48
• /
• 2014

Experimental study of sedimentation and self-weight consolidation has been primary research area in dredged soil. However, good quality of the dredged soil and minimum water pollution caused by the pumping of reclaimed soil require intensive study of the flow characteristics of dredged material due to dumping. In this study, continuity and the equilibrium equations for mass flow assuming single phase was derived to simulate mass flow in dredged containment area. To optimize computation and modeling time for three dimensional geometry and boundary conditions, depth integration is applied to governing equations to consider three dimensional topography of the site. Petrov-Galerkin formulation is applied in spatial discretization of governing equations. Generalized trapezoidal rule is used for time integration, and Newton iteration process approximated the solution. DG and CDG technique were used for weighting matrix in discontinuous test function in dredged flow analysis, and numerical stability was evaluated by performed a square slump simulation. A comparative analysis for numerical methods showed that DG method applied to SU / PG formulation gives minimal pseudo oscillation and reliable numerical results.

• Computational Structural Engineering
• /
• v.11 no.3
• /
• pp.199-212
• /
• 1998

본 연구에서는 기설구조물에 대한 손상도 추정기법과 내진능력평가 방법에 대하여 연구하였다. 구조물의 손상도를 추정하는 방법으로는 소수의 계측 데이터를 이용한 모드섭동법(inverse modal perturbation)을 이용하였다. 구조물의 손상은 강성행렬의 감소로 표현하여, 각 요소행렬에 대한 손상을 손상지수를 사용하여 나타내었다. 구조적 손상과 이에 기인한 고유진동 특성의 변화량과의 관계를 섭동방정식으로부터 구한 후, 이로부터 손상지수와 고유진동 특성의 변화량과의 관계를 유동하였다. 따라서 손상 전과 후에서 구조물의 고유진동수와 모드형상을 측정하여 섭동식의 해를 구함으로써 구조물의 강성행렬의 감소로 나타나는 구조물의 손상도를 추정하게 된다. 손상도 추정에 의해 평가된 강성의 변화량에 기인한 손상 후의 기설구조물의 지진응답, 내진능력과 지진손상도의 평가를 손상전과 비교하였다. 내진능력은 구조부재에서 회전연성도 능력의 경험식을 이용하여 평가하였고, 지진손상도의 평가는 가장 많이 사용되는 방법인 Park & Ang 방법을 사용하였다. 예제해석은 다른 지진하중을 받는 2층과 8층의 예제구조물에 대해서 수행하였다.

• 한국전산유체공학회:학술대회논문집
• /
• 2005.04a
• /
• pp.219-222
• /
• 2005

A compressible two-fluid two-phase flow computation model using the stiffened-gas equation of state is formulated. Since the conservation equation system is of mixed type, it gives complex eigenvalues. The sonic speeds obtained from the individual single phase have been simply used in the literature for the fastest wave speeds necessary in the HLL scheme. This method has worked fine but proved to be quite diffusive according to our test. To improve the accuracy, we here propose to utilize the analytic eigenvalues evaluated from an approximate Jacobian matrix lot the fastest wave speeds. The interfacial transfer terms were dropped in constituting the Jacobian matrix for this purpose. The present scheme proved efficient, robust and accurate in comparison with other existing methods. We solved the cavitating flow problem using the present scheme. The result shows more detailed wave structure in the cavitating process caused by the strong expansion waves.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.35 no.3
• /
• pp.183-195
• /
• 2007

Comprehensive mathematical comparison of numerical dissipation vector was made for a compressible and the preconditioned version Roe's Riemann solvers. Choi and Merkle type preconditioning method was selected from the investigation of the convergence characteristics of the various preconditioning methods for the flows over a two-dimensional bump. The investigation suggests a way of migration from a compressible code to a preconditioning code with a minor changes in Eigenvalues while maintaining the same code structure. Von Neumann stability condition and viscous Jacobian were considered additionally to improve the stability and accuracy for the viscous flow analysis. The developed code was validated through the applications to the standard validation problems.

• Transactions of the Korean Society for Noise and Vibration Engineering
• /
• v.16 no.4 s.109
• /
• pp.387-393
• /
• 2006

In this paper, a spectral element model is developed for the uniform straight pipelines conveying internal unsteady fluid flow. The spectral element matrix is formulated by using the exact frequency-domain solutions of the pipe-dynamics equations. The spectral element dynamic analysis is then conducted to evaluate the accuracy of the present spectral element model and to investigate the vibration characteristics and internal fluid characteristics of an example pipeline system.