Convolutional neural network has made great breakthroughs in the field of computer vision. With its powerful image processing ability,we transform classification of underground sedimentary salt-body into image semantics segmentation problem. Deep convolution neural network is applied to implement pixel-level semantics segmentation of salt seismic images. We increase depth of the network and adds batch normalization and Dropout processing based on U-Net, which makes the neural network model with higher reliability and stronger generalization ability. With experiments, it was found that adding batch normalization layer after convolution layer, and adding Dropout after the pooling layer and concatenate layer improve steadily segmentation performance of the model for salt-body seismic image.

To investigate entropy generation caused by a grooved structure on flow, we used Fluent software and large eddy simulation (LES) technology to analyze entropy generation in flow over a plate with a grooved structure. It showed that a grooved structure was able to reduce entropy generation in the flow. At 30 m·s^-1(Re≈20 000), total entropy generation in near-wall region decreased by approximately 25%. And at 40 m·s^-1(Re≈27 000), it decreased by approximately 19%. However, the grooved structure did not have a sustained effect on subsequent flow. As distance from the wall in the normal direction increased, the effect decreased gradually. Entropy production at the top of the trench is much greater. And entropy production caused by turbulence is dominant. It provides a theoretical basis for optimizing drag reduction of trench structures.

Particle swarm optimization (PSO) algorithm has strong ability to explore global optimal region for heat exchanger networks synthesis. However, particles may trap into local optima and converge prematurely in late evolution. Therefore, an improved particle swarm optimization algorithm based on diversity feedback and real-time updating strategy is proposed. Firstly, index of population health degree is established to evaluate population diversity during evolution. Secondly, a random perturbation strategy and a centrifugal strategy are combined respectively with PSO algorithm to enrich population diversity and enhance global search ability. Furthermore, gradient search strategy is applied to search efficiently local optima and improve computational efficiency of PSO algorithm. Finally, a feedback mechanism of population health degree is proposed to real-time monitor health status of population and further to adopt different update strategies for keeping particles healthy during evolution. The method was applied to several cases taken from literature and results are encouraging. They are better than those of other improvements for PSO.

Take use of characteristics of inertial confinement fusion(ICF) models and associated numerical discrete schemes, we proposed multiple techniques for promoting computational efficiency of two-dimensional radiation-hydrodynamic code LARED-integration. Numerical results show that with unchanged parallel resources it is two times faster in downstream spatial sweeping of solving transport equations. Half of total running time is saved for integrated simulation of laser fusion experimental hohlraums.

In practical applications, equation of states (EOS) consists of several piecewise smooth surfaces, which leads to discontinuity at interface. As a traditional nonlinear iterative algorithm is applied to an energy equation with discontinuous EOS, it may lead to slow convergence and unphysical solutions. To overcome the difficulties, a nonlinear problem is designed, and a nonlinear iterative algorithm is proposed to solve the problem. The algorithm is fit for energy equations with discontinuous EOS of piecewise smooth functions. A parameter of energy change is defined in the algorithm so that it is unnecessary to know discontinuity position in advance. The algorithm calculates precisely net gain or leakage of energy, which can be used to assess influence of discontinuity in EOS. Typical numerical experiments verify that the algorithm converges stably, and gives physical solutions.

We analyze non-physical solutions in laser ray tracing simulation.With a system of nonlinear equations,a new algorithm for intersection is presented.A special preconditioner is used to improve numerical stability.The algorithm is suitable for any intersectant condition.Numerical experiments show that the algorithm has better precision and adaptability.Simulation in LARED code shows good performance.

Convergence of source iteration is analyzed for multi-group radiative transfer calculations. A two-level nesting transport acceleration method is developed. Numerical results show speedup factors of the scheme are higher than that of GTA method. The scheme is feasible for non-rectangle grid calculations of 2D problems with large discontinuities of material properties.

We discuss deterministic numerical methods such as discrete ordinates methods,spherical harmonic dimensional and iteration acceleration method,for particle (neutron and radiation) transport equations in one-dimensional spheric geometry and twodimesional cylindrical geometry. Recent advances on numerical methods of transport problems are briefly described,including adaptive time step discrete scheme,iterative methods for eigenvalue problems,modified subcell balance methods,simplified spherical harmonic methods,simulation methods for coupling of diffusion and transport,parallel discrete scheme with interface prediction and correction,grey transport synthetic acceleration method,etc. Based on the analysis of difficulties in numerical methods for transport equations,suggestions for future work are proposed.

Energy levels of ground state and low-lying excited states of hydrogenic impurities in InAs quantum rings are investigated with effective mass approximation and perturbation method.It shows that energy levels of electron do not change with electronic radial coordinate.They have characteristics of two-dimensional hydrogen atom energy levels in parabolic potential platform area.Energy levels of electron are sensitively dependent on radius of the quantum ring.There exist a minimum on account of parabolic confinement potential in area of parabolic potential with finite depth.Degenerate energy levels of the first excited state for hydrogenic impurities are not relieved.As n ≥ 2 degenerate energy levels are split and energy spacings increase with increase of the radius.

In inertial confinement fusion (ICF) target, collision between fusion neutrons may produce super-high energy neutrons. A relationship between velocity of fusion and ratio of generated super-high energy neutron is derived. An ideal fusion model is simulated by numerical method. It indicates that the ratio of generated super-high energy neutron is increased sharply as burning volume of DT is decreased. With this information, 2D-effect and mixing-effect during compression of target are deduced with escaped super-high energy neutrons. Furthermore, a relationship between velocity of burning T and ratio of generated super-high energy neutron is concluded. It shows that using super-high energy neutron to detect ICF is effective.

An adding weight sample method is developed for neutron transportation in inertial confinement fusion.It aims to increase samples in important zone and decrease samples in non-important zone.Weight of each sample is corrected for keeping calculated results non-bias.Typical model is simulated.It indicates that the method improves effectively neutron samples in fusion zone,while collisions between neutrons increase obviously.Calculation error of super-high energy neutron flux is decreased remarkably.Calculation efficiency of super-high energy neutrons is improved.

Chemisorption of energetic C₂ clusters on diamond(111) surface is investigated with molecular dynamics simulation and mang-body interatomic Brenner(#2) potential.At 300 K,molecular dynamics simulation are performed as a C₂ cluster bombards six different bombarding sits on diamond(111) surface at incident energies of 1 eV,20 eV,30 eV,respectively.Different chemisorption configurations are observed.Deposition of C₂ atoms are observed during the bonding collision.Influence of sits and impact energy on structure of deposited clusters are discussed.It shows that the collision processes change in different environments.Increase of cluster incident energy is apt to bond forming.

Relationship between super-high energy neutrons and collisions is derived.Monte Carlo method is used to decide speed of neutrons after collision.A code applicable to nonlinear neutron transport problem is developed to simulate neutron collision and transportation.Numerical calculation indicates that production rate of super-high energy neutrons is in linear relationship with the square of neutron source intensity.

Molecular dynamics (MD) simulation is made to study growth of diamond-like carbon (DLC) films via ion-beam-assisted deposition (IBAD).C₂ molecules and H ions are selected as deposition projectiles and Ar ions are selected as assistance projectiles.At fixed incident energy and varied ration (Ar/C),assisted deposition of film is investigated.Transient mobility and migration of C atoms due to Ar impact is investigated.It shows that the impact-induced high recoil energy and displacement of deposited C atoms play a key role in growth of DLC films.It is attributed to incident energy and momentum of assistance Ar.The results agree well with experimental observation.It helps understanding of the growth mechanism.

We present an improved relaxation algorithm involving rotation of non-spherical particles.In a sphere assembly model,the algorithm simulates random packing of non-spherical particle of any shape as well as particle mixture consisted of particles with different shapes.In the sphere assembly model,particle shape is described with contour of a number of spheres.Contacts between non-spherical particles are treated as contacts of spheres.By introducing torque and rotation of non-spherical particles,the algorithm overcomes self-lock problem,and random close packing is achieved.Packing densities obtained are coincident with numerical and experimental results in literatures.

Strong-coupled multigroup radiative transfer in optically thick regions are analyzed.A simple conner balance method in 2D is presented.A grey transport acceleration scheme is generalized to accelerate source iteration convergence of differenced multigroup transport equations.

An approximate linearization method is proposed for nonlinear neutron transport equations. The equation is deduced to a form suitable for Monte Carlo simulations. Numerical results show that it is reasonable. It provides a tool for simulating high energy neutron transportation.

A consistent linear multifrequency-grey acceleration scheme is developed to accelerate the iterative convergence speed of diamonddifferenced multigroup radiative transfer equations. A difference scheme of acceleration equation is obtained directly by applying discrete P1 approximation to differenced S_N equations. Numerical examples show that the method is effective and robust.

A relaxation algorithm is developed to simulate sphere packing with arbitrary diameter distribution.An adaptive iteration period is employed to keep the packing density stable with various number of spheres.The packing density and the coordinate number of the algorithm are higher than those of previous approaches.The efficiency of contact detection is considerably increased by background grids and double link group structure.The time complexity of the algorithm is O(N),where N is the number of spheres.It needs only CPU time of 217s to achieve a packing density of 0.64 for random close packing of 10 000 equal spheres with an AMD Athlon 3200+PC.New classification of numerical methods for sphere packing is also presented.

With shape functions of mean value interpolation and a bivariate Taylor expression,error estimation of mean value interpolation within polygonal elements is analyzed.Inequality of error estimation for mean value interpolation is derived.Mean value interpolation in the approxmation of temperature distributions on a convex domain is shown.

Propagation of light in a supersonic flat turbulence boundary layer is numerically analyzed. The flow characteristics are studied according to the optic wave-front variance. The mass-weight averaging N-S equations are adopted as the governing equations of flow, and a two-equation turbulent model is employed. Light propagation is described by paraxial wave equations and solved by a phase-screen method combined with the FFT technique, Light intensity distribution and phase aberrations are given in order to show the optic aberrations as well as the flow characteristics.

A parallel algorithm for time-independent Monte Carlo transport is successful since particles are independent and they are distributed to multiple processors.However,for time-dependent Monte Carlo transport problems, the parallel efficiency reduces and the parallel scale is limited due to the communication of scattering source attribute and meshes in each time-step.We propose two algorithms in them adaptive processor assignment and optimized processor choice are obtained.With a Monte Carlo stratified sampling technique for scattering source treatment the communication cost is reduced greatly.The parallel expandability is improved.A large speedup over the basic algorithm is obtained.

A 3-D multigroup P₃ approximation Monte Carlo code MCMG-BURN is developed by coupling the neutron transport and burnup.MCMG-BURN code is based on the continuous-energy cross-section Monte Carlo code MCNP and the lattice homogeneous code WIMS.It uses the multigroup cross-section libraries to simulate the critical test reactors and HFETR(High Flux Engineering Test Reactor).The agreement results with the MCNP results and experiments are achieved.The MCMG-BURN code is at almost the same precision with the MCNP code while requiring considerably less computing time.

The Global Positioning System (GPS) ray-shooting method is a self-sufficient observation operator in GPS/MET (Meteorology) data variational assimilation.The huge computation of the GPS ray-shooting method,however,makes it impossible so far to be applied in data assimilation and operational prediction.In order to reduce the huge computation,a 2nd-order timesaving symplectic scheme is used to solve the equations of the GPS ray trajectory due to its separable Hamiltonian nature.Not only does it save 75% of the CPU time the old GPS ray-shooting model using 4th order Runge Kutta method takes,but also does the simulation accuracy be improved slightly to some extent.Then the parallel computing of the new GPS ray shooting model is studied.The speed up and the parallel efficiency are both very high.

It outlines MCMGP3-A 3-D multigroup P₃ Monte Carlo neutron transport code, which is designed for computation of reactor critical safety. It is developed from the coupled neutron and photon transport Monte Carlo code MCNP with continual point cross sections and the continual point cross section module of MCNP code has been replaced by multigroup cross section module. MCMGP3 code is with the capablilities of MCNP geometry treatment, counting, deviation reducing techniques and plot. Either macroscopic or microscopic cross sections can be used.The neutron scattering angular distribution adopts P₃ approximation and generalized Gaussian quadrature scheme. The sample problem results prove the computational correctness of the code. MCMGP3 code can treat steady or unsteady porblems with exterior source and critical problems. It suits various cross section formats and the energy groups can be expanded easily from one group to multigroup. In addition, the MCMGP3 code has successfully realized the coupled with the reactor lattic code WIMS. It can be used to simulate the full reactor problems.

The simulation of a time resolved electron microscope is discussed.Temporal resolution and spatial resolution power are evaluated by computing the motion of transmission eletrons emitted from the sample in the combination of static magnetic field and dynamic electric field.The design methods,including the calculation of magnetic field,dynamic electric magnetic field,electron trajectory tracing and spatial and temporal spread functions,are also involved.The simulation demonstrates that 6 framing images with the size of φ=10mm,temporal resolution of 100ps and spatial resolution of 100Å have been achieved in the time resolved electron microscope.