• 한국멀티미디어학회논문지
• /
• 제21권8호
• /
• pp.969-981
• /
• 2018

Recent advances in machine learning approaches such as deep neural network and reinforcement learning offer significant performance improvements in generating detailed and varied motions in physically simulated virtual environments. The optimization methods are highly attractive because it allows for less understanding of underlying physics or mechanisms even for high-dimensional subtle control problems. In this paper, we propose an efficient learning method for stochastic policy represented as deep neural networks so that agent can generate various energetic motions adaptively to the changes of tasks and states without losing interactivity and robustness. This strategy could be realized by our novel trajectory search method motivated by the trust region policy optimization method. Our value-based trajectory smoothing technique finds stably learnable trajectories without consulting neural network responses directly. This policy is set as a trust region of the artificial neural network, so that it can learn the desired motion quickly.

• Wind and Structures
• /
• 제7권3호
• /
• pp.159-172
• /
• 2004

The study of how buildings affect wind flow is an important part of the research being conducted on urban climate and urban air quality. NJU-UCFM, a standard $k-{\varepsilon}$ turbulence closure model, is presented and is used to simulate how the following affect wind flow characteristics: (1) an isolated building, (2) urban canyons, (3) an irregular shaped building cluster, and (4) a real urban neighborhood. The numerical results are compared with previous researchers' results and with wind tunnel experiment results. It is demonstrated that the geometries and the distribution of urban buildings affect airflow greatly, and some examples of this include a changing of the vortices behind buildings and a "channeling effect". Although the mean air flows are well simulated by the standard $k-{\varepsilon}$ models, it is important to pay attention to certain discrepancies when results from the standard $k-{\varepsilon}$ models are used in design or policy decisions: The standard $k-{\varepsilon}$ model may overestimate the turbulence energy near the frontal side of buildings, may underestimate the range of high turbulence energy in urban areas, and may omit some important information (such as the reverse air flows above the building roofs). In ideal inflow conditions, the effects of the heights of buildings may be underestimated, when compared with field observations.

• 천문학회보
• /
• 제39권2호
• /
• pp.74.2-74.2
• /
• 2014

Magnetohydrodynamics(MHD) dynamo depends on many factors such as viscosity ${\gamma}$, magnetic diffusivity ${\eta}$, magnetic Reynolds number $Re_M$, external driving source, or magnetic Prandtl number $Pr_M$. $Pr_M$, the ratio of ${\gamma}$ to ${\eta}$ (for example, galaxy ${\sim}10^{14}$), plays an important role in small scale dynamo. With the high PrM, conductivity effect becomes very important in small scale regime between the viscous scale ($k_{\gamma}{\sim}Re^{3/4}k_fk_f$:forcing scale) and resistivity scale ($k_{\eta}{\sim}PrM^{1/2}k_{\gamma}$). Since ${\eta}$ is very small, the balance of local energy transport due to the advection term and nonlocal energy transfer decides the magnetic energy spectra. Beyond the viscous scale, the stretched magnetic field (magnetic tension in Lorentz force) transfers the magnetic energy, which is originally from the kinetic energy, back to the kinetic eddies leading to the extension of the viscous scale. This repeated process eventually decides the energy spectrum of the coupled momentum and magnetic induction equation. However, the evolving profile does not follow Kolmogorov's -3/5 law. The spectra of EV (${\sim}k^{-4}$) and EM (${\sim}k^0$ or $k^{-1}$) in high $Pr_M$ have been reported, but our recent simulation results show a little different scaling law ($E_V{\sim}k^{-3}-k^{-4}$, $EM{\sim}k^{-1/2}-k^{-1}$). We show the results and explain the reason.

• 제어로봇시스템학회논문지
• /
• 제20권9호
• /
• pp.964-970
• /
• 2014

In this study we have developed an anthropomorphic robot hand and arm by using tendon-tubes which can be used for people's everyday life as a robot's dynamic power transmission device. Most previous robot hands or arms had critical problem on dynamic optimization due to heavy weight of power transmission parts which placed on robot's finger area or arm area. In order to resolve this problem we designed light-weighted robot hand and arm by using tendon-tubes which were consisted of many articulations and links just like human's hand and arm. The most prominent property of this robot hand and arm is reduction of the weight of robot's power transmission part. Reduction of weight of robot's power transmission parts will allow us to develop energy saving and past moving robot hands and arms which can be used for artificial arms. As a first step for real development in this study we showed structural design and demonstration of simulation of possibility of a robot hand and arm by tendon-tube. In the future research we are planning to verify practicality of the robot hand and arm by applying sensing and controlling method to a specimen.

• 천문학회보
• /
• 제37권1호
• /
• pp.92.2-92.2
• /
• 2012

Previous simulations posed a problem that they used reduced ion to electron mass ratios to save computation time. It was assumed that ion and electron dynamics are sufficiently separated, but it was not clearly verified. In this study, we examine the effect of ion to electron mass ratios on the generation of ion holes by ion beam driven instability. Ion holes are generated via electron holes in an applied electric field with the given initial condition. First, the ion acoustic instability is excited and nonlinearly develops. After the ion acoustic instability nonlinearly develops, the ion two-stream instability is excited and develops into ion holes. This implies that the previously suggested ion beam driven instability is strongly affected by the coupling between ions and electrons and the ion to electron mass ratio is important on the development of the instability. The energy transition and detail variation is different as reduced mass ratio under the same observation value based on FAST satellite. Although, the parameters are rescaled by conserving the kinetic energy to obtain the proper results, the nonlinear evolution is not perfectly identical.

• 천문학회보
• /
• 제42권2호
• /
• pp.46.1-46.1
• /
• 2017

Late-type galaxies falling into a cluster would evolve being influenced by the interactions with both the cluster and the nearby cluster member galaxies. Most numerical studies, however, tend to focus on the effects of the former with little work done on those of the later. We thus perform numerical study on the evolution of a late-type galaxy falling radially toward the cluster center interacting with neighbouring early-type galaxies, using N-body, hydrodynamical simulations. Based on the information about the typical galaxy encounters obtained by using the galaxy catalog of Coma cluster, we run the simulations for the cases where a Milky Way Galaxy-like late-type galaxy, flying either edge-on or face-on, experiences six consecutive collisions with twice more massive early-type galaxies having hot gas in their halos. Our simulations show that the evolution of the late-type galaxy can be significantly affected by the high-speed multiple collisions with the early-type galaxies, such as on the cold gas content and the star formation activity, particularly through the hydrodynamic interactions between the cold disk and the hot gas halos. By comparing our simulation results with those of others, we claim that the role of the galaxy-galaxy interactions on the evolution of late-type galaxies in clusters could be comparable with that of the galaxy-cluster interactions, depending on the dynamical history.

• 한국의학물리학회지:의학물리
• /
• 제6권1호
• /
• pp.65-77
• /
• 1995

A new gamma camera using a-Si : H photodetectors has been designed for the imaging of heart and other small organs. In this new design the photomultiplier tubes and the position sensing circuitry are replaced by 2-D array of a-Si : H p-i-n pixel photode tectors and readout circuitry which are built on a substrate. Without the photomultiplier tubes this camera is light weight, hence can be made portable. To predict the characteristics and the performance of this new gamma camera we did Monte Carlo simulations. In the simulations 128${\times}$128 imaging array of various pixel sixes were used. $\^$99m/Tc(140keV)and $\^$201/Tl(70keV) were used as radiation sources. From the simulations we could obtain the resolution of the camera and ther overall system, and the blurring effects due to scattering in the phantom. Using the Wiener filter for image processing, restoration of the blurred image could be achieved. Simulation results of a-Si : H based gamma camera were compared with those of a conwentional gamma camera.

• 한국의학물리학회지:의학물리
• /
• 제27권4호
• /
• pp.236-240
• /
• 2016

The purpose of this study is to propose a novel high sensitivity neuro-PET design. The improvement of sensitivity in neuro-PET is important because it can reduce scan time and/or radiation dose. In this study, we proposed a novel PET detector design that combined conical shape detector with cylindrical one to obtain high sensitivity. The sensitivity as a function of the oblique angle and the ratio of the conical to cylindrical portion was estimated to optimize the design of brain PET using Monte Carlo simulation tool, GATE. An axial sensitivity and misplacement rate by penetration of ${\gamma}$ rays were also estimated to evaluate the performance of the proposed PET. The sensitivity was improved by 36% at the center of axial FOV. This value was similar to the calculated value. The misplacement rate of conical shaped PET was about 5% higher than the conventional PET. The results of this study demonstrated the conical detector proposed in this study could provide subsequent improvement in sensitivity which could allow to design high sensitivity PET for brain imaging.

• 한국진공학회:학술대회논문집
• /
• 한국진공학회 2014년도 제46회 동계 정기학술대회 초록집
• /
• pp.234.1-234.1
• /
• 2014

Many studies have been investigated on high density plasma source (Electron Cyclotron Resonance[ECR], Inductively Coupled Plasma[ICP], Helicon plasma) for large area source after It is announced that productivity of plasma process depends on plasma density. Among them, Some researchers have been studied on multiple sources In this study, we attempted to determine the possibility of multiple inductively coupled plasma (ICP), and helicon plasma sources for large-area processes. Experiments were performed with the one and two coils to measure plasma and electrical parameters, and a circuit simulation was performed to measure the current at each coil in the 2-coil experiment. Based on the result, we could determine the possibility of multiple ICP sources due to a direct change of impedance due to current and saturation of impedance due to the skin-depth effect. However, a helicon plasma source is difficult to adapt to the multiple sources due to the consistent change of real impedance due to mode transition and the low uniformity of the B-field confinement. As a result, it is expected that ICP can be adapted to multiple source for large-area processes.

• 한국연소학회:학술대회논문집
• /
• 한국연소학회 2006년도 제32회 KOSCO SYMPOSIUM 논문집
• /
• pp.213-217
• /
• 2006

Recent advances in energetic materials modeling and high-resolution hydrocode simulation enable enhanced computational analysis of bio-medical treatments that utilize high-pressure shock waves. Of particular interest is in designing devices that use such technology in medical treatments. For example, the generated micro shock waves with peak pressure on orders of 10 GPa can be used for treatments such as kidney stone removal, trans-dermal micro-particle delivery. and cancer cell removal. In this work, we present a new computational methodology for applying the high explosive dynamics to bio-medical treatments by making use of high pressure shock physics and multi-material wave interactions. The preliminary calculations conducted by the in-house code, GIBBS2D, captures various features that are observed from the actual experiments under the similar test conditions. We expect to gain novel insights in applying explosive shock wave physics to the bio-medical science involving drug injection. Our forthcoming papers will illustrate the quantitative comparison of the modeled results against the experimental data.