The element-free Galerkin method (EFGM), which is one of the meshless method, has been tried to be applied to the stress concentration, crack and shells problems. Using the moving least square approximation (MLSM) as the interpolation function is a special feature for the EFGM. Another feature for the EFGM is that the EFGM has the continuity of the first derivative i.e. strain and stress for a structural analysis by selecting the weight function. Then we can obtain displacement, strain and stress anywhere. Calculating the fracture mechanics parameter, we can calculate more accurate fracture mechanics parameter for nonlinear fracture mechanics problems. Therefore, the EFGM is a promising method for treating problems such as crack growth because it can easily obtain the variations of the fracture mechanics parameters. The EFGM has not been applied to nonlinear problems such as elastic-plastic problems. If the EFGM will be able to applied to the nonlinear problems, it will be useful as a numerical method for solid mechanics. In the present paper, the EFGM is applied to elastic-plastic problems and to estimating J-integral. The results obtained from the EFGM agree well with those of the finite element method, and the EFGM is successfully applied to the elastic-plastic fracture mechanics problem.

View full abstract

This paper presents the first coupling application of the dual reciprocity BEM (DRBEM) and dynamic programming filter to inverse elastodynamic problem. The DRBEM is the only BEM method, which does not require domain discretization for general linear and nonlinear dynamic problems. Since the size of numerical discretization system has a great effect on the computing effort of recursive or iterative calculations of inverse analysis, the intrinsic boundary-only merit of the DRBEM causes a considerable computational saving. On the other hand, the strengths of the dynamic programming filter lie in its mathematical simplicity, easy to program and great flexibility in the type, number and locations of measurements and unknown inputs. The combination of these two techniques is therefore very attractive for the solution of practical inverse problems. In this study, the spatial and temporal partial derivatives of the governing equation are respectively discretized first by the DRBEM and the precise integration method, and then, by using dynamic programming with regularization, dynamic load is estimated based on noisy measurements of velocity and displacement at very few locations. Numerical experiments involved with the periodic and Heaviside impact load are conducted to demonstrate the applicability, efficiency and simplicity of this strategy. The affect of noise level, regularization parameter, and measurement types on the estimation is also investigated.

View full abstract

The application of the lattice BGK Poisson solver to the quenched KPZ (QKPZ) equation, which describes fluid interfaces in porous media, is presented. The lattice BGK QKPZ solver is the new convenient tool for studying growing interfaces by numerical simulations. Using this new solver in 1+1 dimensions, we confirm the existence of two universality classes by introducing some assumptions to the QKPZ nonlinear term and discuss the third universality class for the QKPZ equation depending on the QKPZ nonlinearity. We also compare our simulations with experimental results, showing that our QKPZ solver can be applied to analyze and predict growth phenomena in porous media. The lattice BGK method has many applications in numerical simulations of various types of fluid flows such as the fluid growth in disordered media.

View full abstract

In this paper, an idea named LFD(Leg Functions Distribution) for a quadruped robot is explained. A quadruped robot which reflects this idea has different leg functions between fore-feet and hind-feet. The fore-feet have the gravitational support function. Conventional walking robot has the center of gravity at the center of the fuselage, however this machine has the center of gravity at the front of the fuselage. The hind-feet have the function of generating driving forward force. Because the center of gravity is distributed at the front part, it makes possible for the hind-feet to reduce generating the gravitational support force. This simple mechanism is developed by using a counter balance. In this paper, V/H ratio is also defined. V/H ratio shows the proportional value in percent of this idea. Some numerical results using DME(Dynamic Manipulability Ellipsoid) and V/H ratio are described.

View full abstract

The finite deformation kinematics for two-scale modeling of heterogeneous media is originally described in this paper based on our own developments of the multi-scale computational strategy. With the help of three-field variational formulation and the convergence results, we derive the kinematically correct formulation for two-scale finite deformation problems and provide a clear distinction between ours and the ones with two-scale asymptotic expansions with rate formulation. Then the tangential homogenization process is naturally arises as a consistent linearization in the context of two-scale modeling. During the course of this development, the notion of two-scale kinematics is required to properly describe the micro-macro coupling behaviors inherent to this kind of modeling strategies. By using a simple numerical example, we explain our basic idea toward the mathematical modeling based on the homogenization theory.

View full abstract

The Cauchy problem of the Poisson equation in two spatial dimensions is considered in this paper, in which Dirichlet and Neumann data are simultaneously imposed on a part of the boundary of the domain. The problem can be regarded as a boundary inverse problem, in which the proper boundary conditions are to be indentified for the rest of the boundary. The problem can be reformulated as a minimization problem of a functional with constraints, which is minimized by the method of the steepest descent. The minimization problem is recast into successive solution of primary and dual boundary value problems for the Poisson equations. Some examples in this paper indicate that the numerical solutions are convergent and the scheme is confident.

View full abstract

In this paper, Finite Element Method (FEM) is proposed for application to a control system of link mechanisms. The control software using FEM can deal with a sudden change in hardware, and is capable of expressing lack or disability of constituent members of the system only by changing input data since the entire system is subdivided into discrete elements and evaluated as a continuum. It handles the information of the entire system in parallel, which is useful especially when controlling hyper-redundant manipulators. The real-time control by using FEM became possible by applying the Shifted Integration technique, which produces the higher computational efficiency in the finite element analyses of framed structures including static and dynamic collapse problems. A single link structure of a pin joint and a rigid bar is expressed by shifting the numerical integration point in two linear Timoshenko beam elements. This paper describes the modeling of link mechanisms by using the Shifted Integration technique, and a numerical scheme to obtain joint torque curves in n-link mechanism based upon Cartesian coordinates is derived. A simulation test on 8-link mechanism by FEM is carried out and the joint torque curves are compared to those obtained by conventional Newton-Euler method.

View full abstract

This paper describes a draft of logistics planning based on a simulation of traffic flow by Cellular Automata. Cellular Automata is an effective method of emergent computations called Constructive Approach. Traffic flow is a typical example of Complex Systems in which the whole phenomenon from the local interactions of components of the system emerges, therefore logistics planning should be essentially designed considering implications of traffic flow. In this study, traffic flow is modeled by Cellular Automata. Logistics planning is framed by optimizing vehicle allocation plan, vehicle routing plan and so forth with the help of Genetic Algorithm so that the cost calculated from conditions on utilization of trucks considering traffic flow by Cellular Automata might be minimized. The applicability and effectiveness of the proposed approach for logistics planning is investigated with simulation experiments.

View full abstract

In this paper, an effective method for the topology optimization of 3D structures is presented. In this method, the structures are analyzed using the voxel finite element method with CG solver. The density approach is adopted as the topology optimization method in order to analyze the 3D topology with fine mesh. The optimality criteria method is used for solving the optimization problem, and the filtering method is used for preventing the checkerboards in the solution. The effectiveness of the present method is demonstrated by some numerical examples.

View full abstract

Determination of the effect of the process parameters on the final forming quality is very difficult in sheet metal forming process because forming processes experience very complicated deformation. These process parameters have to be determined for the optimum forming condition before the process design. In this study, finite element program is developed for analyzing the deformation in sheet metal forming operation. Design optimization concept is introduced to calculate the process parameters such as the bead force within the given design requirement. The effectiveness of this theory is examined by solving the deep drawing processes.

View full abstract

In this paper, the asymmetric elastic buckling-load carrying capacity of orthotropic cylindrical shells, under axial loading, is discussed. Based upon the reduced stiffness analytical concept, a simple criterion for this imperfection-sensitive buckling problem has been presented. This not only contributes to our understanding of buckling problems, but also supports to rational design for the orthotropic material.

View full abstract

Many methods have been developed for the analysis of the dynamic behavior of piles using simplified models where either non-linear behavior of the pile near field were neglected or a full non-linear analysis were conducted with considerable computational efforts. The proposed method has the objective of combining the merits of the above-mentioned methods. It presents a new numerical approach to analyze the pile-soil interaction system taking into account the effects of the behavior of the soil at infinity and at the same time accounting for the non-linearities existing in the soil adjacent to the pile, hence the time domain analysis.

View full abstract

The noise prediction computer system (“this system”) can support the anti-noise measure planning of large plants such as thermal power stations. For getting a better planning in a limited term, high speed computing of many case studies is required. The auther parallelized this system by two methods. The first method, very simple one, is effective on a PC cluster composed of the same PCs. If a PC cluster is composed of varied PCs, you can make each PC do his work with little idling time by the second method, a little complicated one. This paper describes these two methods’ algorithms and their parformance evaluations.

View full abstract

In this paper, bucket-by-bucket mesh generation technique is applied to three dimensional geometry and parallel environment. This mesh generation technique is developed for extremely large scale finite element analysis and easily implemented on parallel environment. In this method, the analysis domain is decomposed spaciously to the so-called buckets and mesh generation process is performed on each bucket independently. Hence, parallel computation of this method can be realized only by partitioning buckets to each processor element. Some numerical examples are shown in this paper.

View full abstract

This paper presents an application of the dual reciprocity boundary element method to inverse heat conduction problems. To deal with the ill-posed nature of the problem, iterative regularization methods of conjugate gradient type have been used. Preconditioned biconjugate (PBCG) method has been used for the solution if the discretized boundary element system has as many equations as the total number of unknowns. Preconditioned conjugate gradient (PCG) method has been utilized for the general case in which more equations are available than unknowns. Numerical results show PBCG algorithm has very limited applicability for inverse problems. Further, the PCG algorithm performs better than the PBCG method, and is insensitive to measurement errors. However, for a general inverse problem, the DRBE-PCG algorithm is seen to be sensitive to the location of measurement points, and use of additional regularization of the boundary element system would be required to take care of this aspect.

View full abstract

Iterative domain decomposition method (DDM) is one of effective methods for large scale problems for its excellent parallelism and variety of studies have already been done. As DDM needs iterative calculations between subdomains, it is important to reduce the number of iterations for its speed-up. In this research, we consider the use of Balancing Domain Decomposition (BDD) proposed by Mandel for DDM based on CG method which is adapted to arbitrary domain decomposition.

View full abstract

This paper describes a traffic flow simulator in a parking area of large-scale store and surrounding road network by Cellular Automata. In order to avoid the traffic jam, it is important to investigate the traffic flow of the road network by simulation under various conditions. In the present paper, Cellular Automata was introduced in modeling the traffic flow. The parking area and surrounding road network were divided into cells, on which cars defined as one of state variables moved in relation to other state variables on neighboring cells along discrete time step. The simulation was carried out under several situations of the road network connections and the traffic quantity. As a result, the traffic flow of a road network including parkings could be simulated under various conditions of road and traffic.

View full abstract

In the process of optimization with a large nonlinear dynamic analysis, a large amount of computational cost is required to obtain the optimal design. A large portion of this cost can be avoided using Response Surface Model (RSM) by approximating the more costly analysis. Therefore, the optimization with RSM is studied by various approaches. Sequential Approximate Optimization (SAO) that allows RSM to be updated with new design points during optimization obtains final design better then the method of non updating RSM. But, the design is not always optimal design. In this paper, we propose a method that cross SAO’s design and a good design of current process of optimization by the real type crossover model(RXM). RXM was referred crossover models of Real coded Genetic Algorithms, was developed. We applied a multimodal parameter problem, a cantilever problem and a nonlinear dynamic analysis problem, obtained good results.

View full abstract

This paper presents a PCG method using the Sylvester law of inertia for the eigensolution of large, sparse and symmetric matrices. The proposed method, retaining the advantages of the conjugate gradient method, permits to count the number of sign changes for given matrices by the Sylvester law of inertia, and is able to overcome the numerical difficulty caused in the case where the solution converges to the true eigenvalue. This method is particularly useful to find only small numbers of lower eigenpairs in the large sparse system. The accuracy and stability of this method are confirmed by using several numerical examples. The numerical results give a good agreement even in the systems with multiple eigenvalues.

View full abstract

In this study, we intend to construct MD (Molecular Dynamics) simulation system, which can proceed huge scale MD simulation with real time visualization. As the parallel computer, the workstation cluster can be used by MPI (Message Passing Interface). In this system, there is no extra data transfer and no special computer for the visualization, since the data in the each WS (work station) is visualized by itself. Observers can observe the detail by focusing to the each WS and, at the same time, they can observe whole region of the simulation by looking around. The efficiency of the present system is confirmed by some MD simulations.

View full abstract

GeoFEM is solid earth simulator software, which is under development to be used by super parallel computer, “Earth Simulator” (GS40). The GeoFEM system is currently open to the public as a parallel FEM analysis code. Large-scaled static linear problem up to 100M (100,000,000) degrees of freedom was already analyzed and it shows high performance computing ability by parallel processing. This paper describes a performance validation of the GeoFEM to get a guideline of optimization for the Earth Simulator. As the result, the GeoFEM shows high parallel performance on distributed memory type parallel computer with scalar processor, but the single PE speed is relatively slow. It is assumed that the parallel performance does not drop so much, when we try to speed-up the GeoFEM code by vector tuning or optimization to improve the single node performance

View full abstract

In this paper, new discrete method which applied the approach of the hybrid model to a seepage flow problem is developed. First, the formulation which introduced the potential continuity by penalty in general weighting resudial procedure is shown. Compatibility of the potential on the element boundary edge is approximately introduced using the penalty. Next, the Taylor’s series expansion is used for potential function. It is possible that this model uses the element of the arbitrary shape, because independent potential field is assumed for each element. The equivalent solution with the analytical solution when it examined the accuracy of numerical solution using this model is obtained.

View full abstract

The structural identification and dynamic control of large-scale structures are considered to be difficult due to the structural complicacy and system uncertainties. Active Mass Driver(AMD) has been used as an efficient control actuator based on conventional control methods and control design. In this paper, based on the concept of decentralized information structures for large scale systems and artificial neural networks, a decentralized non-parametric identification method which is the basement for earthquake response control of large-scale structures by AMD system is proposed. And because accelerometers can readily provide reliable and inexpensive measurement of absolute structural acceleration at strategic points on a structure, development of identification method based on acceleration feedback is presented. The effectiveness of the decentralized identification is evaluated through numerical simulations. It is shown that the decentralized neural-emulators enable to identify the coupled system and to reproduce the structural response under different seismic excitations with a great accuracy.

View full abstract

It has been problematic to develop simple and practical automatic mesh generation techniques that can minimize the amount of required input data and can generate a lot of nodes. In particular for the application of automatic mesh generation to stress concentration area simple computer algorithms for managing input and output data are necessitated. In order to satisfy these requirements the Fuzzy reasoning is chosen here as an automatic mesh generation scheme which is applicable to the triangular element. The application target of this method is to accurately evaluate the stress concentration factor of the plate with hole. The element refinement algorithm are used here to optimize meshes based on the element shape and to smooth strees obtained from the Fuzzy reasoning.

View full abstract

The purpose of this paper is to consider a numerical method for the solution of unbounded domain problems of the Poisson equation. An artificial boundary decomposes the external domain Ω into a bounded subdomain Ω₀ and an external subdomain Ω₁ outside Ω₀. The boundary value problems in domains Ω₀ and Ω₁ are indirectly combined by the Dirichlet-Neumann map on the artificial boundary. The finite element and the boundary element methods are applied to the boundary value problems in domains Ω₀ and Ω₁ respectively. By numerical computations, numerical properties of our method are examined. Starting from any initial guess of data on the artificial boundary, the method converges to the exact solution.

View full abstract

Many meshless methods have been proposed to release the formidable task of mesh generation of traditional FEM. However, there yet exit some problems on the way of improving the practicability of meshless methods. Among them, the linear independence of approximation functions should be guaranteed easily, otherwise the nodes have to be distributed with extreme attention to linear independence and the freedom of meshless methods will be limited very much. In this paper, a new meshless approximation method, Cover Least Square Approximation, is proposed. The approximation is constructed on the Finite Cover System by approaching the cover functions as closely as possible. Some properties of Cover Least Square Approximation, such as continuity, interpolatory, reproducibility and linear independence, are discussed. Since the linear independence conditions can be simply satisfied, it can be practicably implemented.

View full abstract

In large scale computational mechanics system, huge data are transmitted among many modules such as mesh generators, domain decomposers, parallel solvers, visualizers, and optimization modules. If the domain decomposition techniques are used in the parallel solvers, the huge mesh data created by the mesh generator, and various attributes attached on the mesh have to be subdivided into subdomain data by the domain decomposer. The data are huge, flows of the data are complicated, and various kinds of data are handled. Therefore, the input/output (I/O) data formats used in the conventional general-purpose finite element codes are not suitable for the system. A new I/O data format system is presented in this paper, which can efficiently handle the various huge data in the large scale parallel computations.

View full abstract

Currently, the finite element method (FEM) is widely implemented in the numerical analysis of structure. However, the creation of the mesh data has become a very burdensome process for today’s analysis which require large-scale complex models. Therefore, the meshless, or element-free, methods have attracted attention, because they require no such mesh, but only nodes. Among the meshless methods proposed, element-free Galerkin method (EFGM) has been considered as the most realistic and practical scheme. However, it has been known that EFG analysis require greater amount of time due to the implementation of the moving least-squares approximate. Therefore, the authors developed a hybrid analysis system which implements both the FEM and EFGM within the same domain. The scheme was applied to two-dimensional linear elasto static problem and its numerical characteristrics were evaluated.

View full abstract

Parallel computation should be needed to analyze the large-scale problem. Iterative solver is effective for the large scale parallel computation. The convergence rate of iterative solvers depends on the problem. Therefore, the applicable area of iterative solvers is restricted. If iterative solver converges for the problem, the efficacy of iterative solver is very large. It is very important to research the area in which iterative solver is efficacy and to develop the application technique of iterative solver. We study the convergence rate of iterative solver for more wide range problem like contact problem (penalty parameter), asymmetric matrix (contact with friction) to extend our study of evaluation of convergence rate of iterative solver for material nonlinearity analysis. The applicability of iterative solvers (CG method, Bi-CGSTAB method, GPBi-CG method) with preconditioning (SSOR, diagonal scaling, block factorization, ILU(IC)) has been studied. CG method with SSOR preconditioning is effective for wide range. CG method with block factorization preconditioning is effective for frictionless contact problem (symmetric matrix). Bi-CGSTAB method with block factorization preconditioning is effective for friction contact problem (asymmetric matrix).

View full abstract

Currently, the finite element method (FEM) is used extensively to solve various complex engineering problems. The MPP environments have been implemented in order to carry out large-scale simulations efficiently and accurately. Control parallel programs implementing the message passing libraries are considered to be superior over the data parallel programs written in languages such as the HPF in that they may be used on diverse MPP environments. Furthermore, fine tuning of the programs by the developers is possible which lead to higher parallel efficiency. In this study, in order to develop a parallel analysis code for large-scale computations with high computational efficiency, data structures and methods of communications were newly proposed and developed based on the EBE-FEM. This type of approach is believed to be effective from the stand point of necessary amount of data storage and is also believed to be suited for parallelization.

View full abstract

Formerly a reference pattern of the actuator angle of the leg of a walking robot had to be prepared beforehand, in the control of a robot. In spite of this complicated task, perfect control was possible. However, as the system is complex in such an exact control method, the calculation time becomes too long, and thus, the robot is unable to walk fast. However, we are able to apply the Genetic Algorithms (GA) to a walking robot. It enables the robot to learn the techniques of walking autonomically more smoothly. And so, the way of walking can be selected according to the surrounding environment. Furthermore many actuators and sensors attached to the robot must be integrated. A Three-Legged-walking Robot was experimentally produced for this study. All legs of the robot have the same mechanism. An analysis of dynamic characteristics of the robot was carried out, and a fundamental walking control method was examined. The robot is controlled by genetic algorithm which is made the non-linear state equation in condition of grounding state, the solution is calculated on a computer, and optimum walking pattern is decided by genetic algorithms. It is difficult to get an optimum solution on non-linear state equation, however, genetic algorithm is useful method to get the solution.

View full abstract

Iterative methods are proposed for a finite element approximation of three-dimensional eddy current problems. The methods are based on an iterative method derived from a perturbation problem of the magnetostatic problem. A “cake” model and the TEAM model are considered as numerical examples. In both examples, the BiConjugate Gradient method is applicable for the complex symmetric linear systems arising in each step of the iterative procedures, the iterative methods converge for a rather wide range of the perturbation parameter, and the present results agree with the previous ones by the mixed method.

View full abstract

Hamiltonian particle dynamics (HPD) is a method to analyze various kinds of scale phenomena from a common viewpoint. The method is based on the Lagrangian particle method and the Hamiltonian formulation of the system. Because the formulation handles impartially with both configuration coordinates and canonical momentum, it is suitable for analyzing dynamic systems such as particle dynamics simulations. However, the method is subject to problems in that the invariants of the system are not conserved in a long-time computation. To solve the problems, symplectic time integrators are introduced and the effectiveness is examined by mathematical and numerical analyses.

View full abstract

There are many kinds of phenomena, ranging from microscopic scale to macroscopic scale and many kinds of numerical analysis methods corresponding to scale size have been introduced. This situation makes it difficult to discuss the phenomena including various scale physics from a common point of view. This is that there is no universal description of phenomena of various scale. This paper focus on Hamiltonian description of phenomena and presents a method of analyzing various kinds of scale phenomena from a common viewpoint. The method is based on both the Hamiltonian formulation of a system and the particle methods, and is called Hamiltonian Particle Dynamics (HPD). In this paper, the background and the fundamental concept of HPD are explained and an example of HPD applications, that is, the behavior of incompressible inviscid fluid, is simulated.

View full abstract

This paper shows usage of mixed element in Free Mesh Method(FMM). The original formulation of FMM has a disadvantage of low accuracy because of its limitation of the available element. Some studies for accurate FMM have been done, but they add other difficulties in data handling. In this study, a mixed element is investigated and it is found to have some desirable features, such that no additional nodes are required, all data are defined on the original node locations, and the additional cost of calculation is reasonably small. Some numerical examples are also demonstrated.

View full abstract

In this paper we present new capabilities of visualization and steering of a parallel program design process in a new computer-assisted problem solving environment (PSE) of P-NCAS. The P-NCAS helps users generate parallel simulation programs based on partial-differential equations (PDEs). These capabilities explore possibilities to visualize and steer the parallel program design process. In our previous paper (Trans. of JSCES, paper No. 19980002) we presented a PSE of NCAS system, which generates numerical simulation programs to solve problems based on PDEs. At present P-NCAS supports a domain decomposition in a design of a parallel numerical simulation program, and the domain decomposition is designed or steered by users through a visualization window. After designing the domain decomposition, the parallel program itself is also designed and generated in P-NCAS, and the designed parallel program is visualized and steered by a PAD diagram. In P-NCAS, MPI functions are employed for message passing, and a SPMD (single program multiple data) model is supported. The visualization and steering capabilities provide users a new flexible design possibility of parallel programs.

View full abstract

Parallel distributed genetic algorithms (PDGA) show better performance in addition to linear-speedup than canonical GA (CGA). However, the detailed analysis of the excellent performance of PDGA has not been performed, and it is uncertain that PDGA outperforms CGA for realistic optimum problems. In this paper, PDGA and CGA applied to solve a typical structural optimization problem to investigate the effectiveness of PDGA. The results of the numerical experiments show that PDGA outperforms CGA for a realistic optimum problems. It is found that PDGA maintains the diversity of the individuals with multiple sup-populations, and the building blocks which are growing in the different sub-populations are combined by the migration.

View full abstract

This paper presents a revision of the free-term coefficient in the Helmholtz integral equation for solution of acoustic problems in a semi-infinite medium when an obstacle is sitting on an infinite plane. Both the integral representation of the velocity potential and the Helmholtz integral equation are re-derived with giving correct values for the free-term coefficient when the formulations utilizing either the half-space Green’s function or its approximation are employed.

View full abstract