基于融合压缩感知的Landweber算法研究

为提高Landweber算法的收敛速度,将压缩感知理论引入Landweber算法中,理论上证明了该算法的可行性并用改进的距离驱动进行系统矩阵的建模。实验结果证明:融合算法进行图像重建时,算法的收敛速度及采样时间均优于Landweber算法。并用GPU对重建算法进行加速,大幅度提高了算法的执行速度。     

一种基于ART算法的快速图像重建技术

ART算法是一个不断迭代的图像重建方法,提高该算法的重建速度一直是研究的重要方面。针对ART算法简化权因子重建模型,提出了一种快速网格遍历算法,通过简单的加减法和比较运算,即可确定射束穿过的网格编号。由于权因子在迭代过程中实时计算,节省了大量的存储空间,大大提高了ART算法的重建速度。实验结果表明本文提出的算法非常有效,与传统方法相比,重建速度提高了近10倍。     

四种系统矩阵建模方法的比较

构建了像素驱动、距离驱动、改进的距离驱动和射线驱动四种系统矩阵建模方法,并将其运用于ADMM算法,比较了它们的建模速度、建模精度以及收敛速度,同时比较了它们在有噪声的情况下对噪声的敏感程度。实验表明:距离驱动的建模速度最快;改进的距离驱动的建模精度最高,射线驱动在ADMM算法中的收敛速度最快。     

融合压缩感知—Landweber法图像重建

为提高Landweber算法的收敛速度,将压缩感知理论引入到Landweber算法中,理论上证明了该算法的可行性并用改进的距离驱动进行系统矩阵的建模。实验结果证明:融合算法进行图像重建时,算法的收敛速度及采样时间均优于Landweber算法。并用GPU对重建算法进行加速,大幅度提高了算法的执行速度。     

GPU加速剂量计算中微分卷积/积分算法的实现

微分卷积/积分算法是计算精度较高的光子线剂量计算算法,较长的计算时间限制了该算法在临床上的使用。本文对微分卷积/积分算法中最耗时的部分实现了基于GPU的并行化计算,与基于CPU的计算相比,在Tesla C1060上计算速度提高可达30X-60X。利用γ因子对计算结果的准确性进行了分析,结果显示,无论是均匀水模还是非均匀头模,在单照射野还是多照射野情况下,加速后的结果都与CPU的计算结果有相同的准确性。通过GPU并行加速,微分卷积/积分算法能成为日常的剂量计算算法。     

GPU加速三维特征线方法的研究

三维特征线方法可以精确求解任意几何堆芯的稳态多群中子输运方程,但同时也具有收敛慢、计算时间长的不足,需要研究相应的加速手段.图形处理器(GPU)计算由于具有速度快,能耗低的优点,被认为是未来高性能计算发展的方向之一.研究GPU计算加速三维特征线方法,并将其应用到三维特征线程序TCM中.借助统一计算设备架构(CUDA)的GPU计算,中央处理器(CPU)负责内存分配、有效增殖系数keff和源分布计算等逻辑性强或归约计算的处理,GPU执行特征线射线扫描细网求解细网通量.计算结果表明,经改写后的程序具有良好的加速效果.     

ART算法高质量重建研究

提高重建质量一直是图像重建重要的研究课题.传统ART算法在重建图像中普遍存在椒盐噪声,通过加入松弛因子,不但加快了算法的收敛速度,而且克服了椒盐噪声.利用计算机仿真实验对松弛因子在不同投影数和不同噪声水平下的重建质量进行了对比分析,结果表明,选择适当的松弛因子,并加入一些先验知识,能够极大地改善重建图像质量.     

GPU梯度自适应SART图像重建算法

针对SART迭代重建算法所需投影数据量大且迭代时间长的问题,将TV算法引入SART中,动态调节梯度步长来加速算法的收敛性能,实现用少量投影数据重建出高质量的图像。为加快算法的执行速度,将SART-TV算法在GPU上并行加速。用SART算法和SART-TV算法对Shepp-Logan模型重建,仿真实验表明:在采样数据相同的情况下,SART-TV算法较SART算法可以重建出更好的图像,且基于GPU的SART-TV算法的迭代时间明显低于基于CPU的SART-TV算法。     

代数迭代算法进行CT图像重建的研究

对CT代数迭代重建算法(ART)进行了研究,讨论了射线数目对图像重建效果的影响,并采用平滑方法,改善了重建图像质量,加快了收敛速度,得出了效果满意的工业CT图像。     

波形板干燥器内液滴模拟的GPU加速研究

基于CUDA C,使用GPU模拟了波形板内大量液滴的二维运动。根据GPU硬件特性,优化了模拟算法。在CPU上实现相同算法,对比了单一液滴直径、不同液滴数量下GPU与CPU的计算时间。结果表明:液滴数量越大,GPU并行加速效果越明显,在液滴数量较大时,得到约60倍的加速。用GPU分析了典型液滴直径分布的情况,得到约30倍的加速。     

具有加速因子的OSEM重建算法用于X射线荧光CT研究

有序子集最大期望值算法(Ordered Subsets Expectation Maximization,OSEM)具有较高的图像重建质量和较短的计算时间,已经被应用于内源CT(如SPECT、PET、同步辐射X射线荧光CT)的图像重建中。本文提出了一种具有加速因子的OSEM算法应用于X射线荧光CT的图像重建,通过引入加速因子h来调制校正因子的步长加快OSEM算法的收敛速度,研究了不同加速因子和不同子集数的AOSEM算法对重建图像质量的影响。计算机模拟及实验结果表明,在获得同等质量重建图像的同时,具有加速因子的OSEM算法的重建速度是常规OSEM的两倍。     

An accelerated fusion power development plan

Energy for electricity and transportation is a national issue with worldwide environmental and political implications. The world must have energy options for the next century that are not vulnerable to possible disruption for technical, environmental, public confidence, or other reasons. Growing concerns about the greenhouse effect and the safety of transporting oil may lead to reduced burning of coal and other fossil fuels, and the incidents at Three Mile Island and Chernobyl, as well as nuclear waste storage problems, have eroded public acceptance of nuclear fission. Meeting future world energy needs will require improvements in energy efficiency and conservation. However, the world will soon need new central station power plants and increasing amounts of fuel for the transportation sector. The use of fossil fuels, and possibly even fission power, will very likely be restricted because of environmental, safety, and, eventually, supply considerations. Time is running out for policymakers. New energy technologies cannot be brought to the marketplace overnight. Decades are required to bring a new energy production technology from conception to full market penetration. With the added urgency to mitigate deleterious environmental effects of energy use, policymakers must act decisively now to establish and support vigorous energy technology development programs. The U.S. has invested $8 billion over the past 40 years in fusion research and development. If the U.S. fusion program proceeds according to its present strategy, an additional 40 years, and more money, will be expended before fusion will provide commercial electricity. Such an extended schedule is neither cost-effective nor technically necessary. It is time to launch a national venture to construct and operate a fusion power pilot plant. Such a plant could be operational within 15 years of a national commitment to proceed.Prepared Under Contract for the Agency for Advancement of Fusion Power, Inc., George S. Clemens, President.     

Soft real-time EPICS extensions for fast control: A case study applied to a TCV equilibrium algorithm

For new control systems development, ITER distributes CODAC Core System that is a software package based on Linux RedHat, and includes EPICS (Experimental Physics and Industrial Control System) as software control system solution. EPICS technology is being widely used for implementing control systems in research experiments and it is a very well tested technology, but presents important lacks to meet fast control requirements. To manage and process massive amounts of acquired data, EPICS requires additional functions such as: data block oriented transmissions, links with speed-optimized data buffers and synchronization mechanisms not based on system interruptions. This EPICS limitation turned out clearly during the development of the Fast Plant System Controller Prototype for ITER based on PXIe platform.In this work, we present a solution that, on the one hand, is completely compatible and based on EPCIS technology, and on the other hand, extends EPICS technology for implementing high performance fast control systems with soft-real time characteristics. This development includes components such as: data acquisition, processing, monitoring, data archiving, and data streaming (via network and shared memory). Additionally, it is important to remark that this system is compatible with multiple Graphics Processing Units (GPUs) and is able to integrate MatLab code through MatLab engine connections. It preserves EPICS modularity, enabling system modification or extension with a simple change of configuration, and finally it enables parallelization based on data distribution to different processing components.With the objective of illustrating the presented solution in an actual tokamak application, we have implemented fundamental tokamak equilibrium quantities such as plasma position, Shafranov shift or internal inductance. The algorithms have been parallelized and implemented for its execution on CPU, GPUs and Matlab, and have been tested using actual magnetic data from the TCV tokamak fast control system.     

松弛系数-有序子集-最大似然期望算法在60Co集装箱CT检测中的应用

将松弛系数和迭代间像素值约束引入有序子集一最大似然期望算法中,构造出应用于发射CT中的ROSEM算法引入到透射式工业CT中。不仅加快了图像重建的收敛速度而且改善了重建图像的质量。针对ROSEM、代数迭代算法（ART）和卷积反投影算法（CBP）,分别做了模拟实验和实际汽车CT检测以及集装箱CT检测实验,并分析比较了它们在^60Co集装箱CT检测中的应用。ROSEM算法无论在重建速度还是重建质量上都优于ART算法,其重建图像质量和CBP算法相当。     

An efficient execution of Monte Carlo simulation based on delta-tracking method using GPUs

An efficient execution method for Monte Carlo simulation using graphic processing unit (GPU) is proposed. The delta-tracking method is used since the delta-tracking method can reduce conditional branches and complexity of code implementation, which degrade computational performance on GPUs. In order to improve parallel efficiency in the eigenvalue calculation, generated fission neutrons are recorded using the atomic operation which avoids the data race in GPUs. We also propose a method to efficiently tally neutron flux in a region. The present method is based on the atomic operation and use of fixed-point type number instead of common floating-point type number. The verification calculations using the C5G7 benchmark problem show effectiveness of the proposed numerical algorithms on GPUs through comparison with calculations using central processing units.     

Research on GPU-accelerated algorithm in 3D finite difference neutron diffusion calculation method

In this paper, the adaptability of the neutron diffusion numerical algorithm on GPUs was studied, and a GPU- accelerated multi-group 3D neutron diffusion code based on finite difference method was developed. The IAEA 3D PWR benchmark problem was calculated in the numerical test. The results demonstrate both high efficiency and adequate accuracy of the GPU implementation for neutron diffusion equation.     

锥束CT扫描方式对ART算法图像重建的影响研究

平板探测器锥束CT的扫描方式对ART(Algebraic Reconstruction Techniques)算法图像重建会产生影响.本文用128×128×128三维Shepp-Logan头部模型作为仿真对象,探讨了不同源-扫描中心距离、投影数、缺角度下扫描方式对图像重建的影响.结果表明,源-扫描中心距离与扫描中心-探...