Shinobu Yoshimura

Advanced general-purpose computational mechanics system for large-scale analysis and design

Abstract We have been developing an advanced general-purpose computational mechanics system, named ADVENTURE, which is designed to be able to analyze a model of arbitrary shape with a 10–100 million degrees of freedom (DOFs) mesh, and additionally to enable parametric and nonparametric shape optimization. Domain-decomposition-based parallel algorithms are implemented in pre-processes (domain decomposition), main processes (system matrix assembling and solutions) and post-process (visualization), respectively. The hierarchical domain decomposition method with a pre-conditioned iterative solver (HDDM) is adopted in the main processes as one of the major solution techniques. Module-based architecture of the system with standardized I/O format and libraries are also developed and employed to attain flexibility, portability, extensibility and maintainability of the whole system. This paper describes some key technologies employed in the system, and shows some latest results including elastic stress analysis of a precise three-dimensional (3D) model of a nuclear reactor vessel with a 60 million DOF mesh on Hitachi SR2201 (1024 PEs ) , and nonparametric shape optimization of a support structure for an express way with a one million DOF mesh on a PC cluster (10 PEs ) .

A Large scale finite element analysis using domain decomposition method on a parallel computer

Abstract A parallel finite element analysis based on a domain decomposition technique (DDT) is considered. In the present DDT, an analysis domain is divided into a number of smaller subdomains without overlap. Finite element analyses of the subdomains are performed under the constraint of both displacement continuity and force equivalence among them. The constraint is satisfied through iterative calculations based on either the Uzawa algorithm or the Conjugate Gradient (CG) method. Owing to the iterative algorithm, a large scale finite element analysis can be divided into a number of smaller ones which can be carried out in parallel. The DDT is implemented on a parallel computer network composed of a number of 32-bit microprocessors, transputers. The developed parallel calculation system named the ‘FEM server type system’ involves peculiar features such as network independence and dynamic workload balance. The characteristics of the domain decomposition method such as computational speed and memory requirement are first examined in detail through the finite element calculations of homogeneous or inhomogeneous cracked plate subjected to a tensile load on a single CPU computer. The ‘speedup’ and ‘performance’ features of the FEM server type system are discussed on a parallel computer system composed of up to 16 transputers, with changing network types and domain decompositions. It is clearly demonstrated that the present parallel computing system requires a much smaller amount of computational memory than the conventional finite element method and also that, due to the feature of dynamic workload balancing, high performance (over 90%) is achieved even in a large scale finite element calculation with irregular domain decomposition.

An automated system for simulation and parameter identification of inelastic constitutive models

This paper presents an automated system for parameter identification of inelastic constitutive models. The system can find good approximate parameters for various identification problems under a user-friendly environment. In order to identify parameters efficiently and in a robust manner, an optimisation method is first proposed. The paper then describes the generalisations applied of modelling, simulation and identification for its various identification uses. Finally, a system, which is developed in conjunction with the generalisations, is described. The performances of the proposed optimisation method and the developed system were investigated with actual material data, and their effectiveness was consequently confirmed.

A parallel finite element method with a supercomputer network

Abstract Computer simulations are about to replace experiments in various fields, and the scale of the models to be simulated tend to be extremely large. To perform large-scale finite element analyses, the authors propose the parallel use of multiple supercomputers connected to one another through a highspeed network. In other words, a supercomputer network is regarded as a parallel computer. As a parallel numerical algorithm for the finite element analysis, we adopt the domain decomposition method (DDM) combined with an iterative solver, i.e. the conjugate gradient (CG) method, where a whole analysis domain is fictitiously divided into a number of subdomains without overlapping. Finite element analyses of the subdomains are performed under the constraint of both displacement continuity and force equivalence among subdomains. Such a constraint can be satisfied through iterative calculations such as the CG method. The present DDM-based parallel finite element algorithm is combined with the server-client model for data and processor management to have the workload balanced dynamically between the processors, and is implemented first on an engineering workstation (EWS) network and then on a supercomputer network. The accuracy and parallel performance of the present system are tested using the network composed of various EWSs. Finally, it is demonstrated that the present system implemented on the supercomputer network can solve the three-dimensional elasticity problem of over one million degrees of freedom at an extremely high average speed of 1.74 GFLOPS. §

Quantitative nondestructive evaluation with ultrasonic method using neural networks and computational mechanics

This paper describes an inverse analysis method using hierarchical neural networks and computational mechanics, and its application to the quantitative nondestructive evaluation with the ultrasonic method. The present method consists of three subprocesses. First, by parametrically changing the location and size of a defect hidden in solid, elastic wave propagation in the solid is calculated with the dynamic finite element method. Second, the back-propagation neural network is trained using the calculated relationships between the defect parameters and the dynamic responses of solid surface. Finally, the trained network is utilized to determine appropriate defect parameters from some measured dynamic responses of solid surface. The accuracy and efficiency of the present method are discussed in detail through the identification of size and location of a defect hidden in solid.

OPTIMIZATION OF A LARGE NUMBER OF COOLANT PASSAGES LOCATED CLOSE TO THE SURFACE OF A TURBINE BLADE

A constrained optimization of locations and discrete radii of a large number of small circular cross-section straight-through coolant flow passages in internally cooled gas turbine vane was developed. The objective of the optimization was minimization of the integrated surface heat flux penetrating the airfoil thus indirectly minimizing the amount of coolant needed for the removal of this heat. Constraints were that the maximum temperature of any point in the vane is less than the maximum specified value and that the distances between any two holes or between any hole and the airfoil surface are greater than the minimum specified value. A configuration with maximum of 30 passages was considered. The presence of external hot gas and internal coolant was approximated by using convection boundary conditions for the heat conduction analysis. A parallel three-dimensional thermoelasticity finite element analysis (FEA) code from the ADVENTURE project at University of Tokyo was used to perform automatic thermal analysis of different vane configurations. A robust semi-stochastic constrained optimizer and a parallel genetic algorithm (PGA) were used to solve this problem using an inexpensive distributed memory parallel computer. 1 Research scientist. ASME member.

Elastic–plastic analysis of nuclear structures with millions of DOFs using the hierarchical domain decomposition method

Abstract The hierarchical domain decomposition method (HDDM) proposed by Comp. Sys. Eng. 4 (1993) 495 is applied to the large scale elastic–plastic finite element (FE) analysis of nuclear structures. The HDDM is a method to implement the finite element method (FEM) on various kinds of parallel environments. The substructure-based iterative methods can effectively be used with the HDDM to solve the large scale linear algebraic equations derived from the implicit FEM. In this paper, some key techniques to parallelize the static elastic–plastic FE analysis by the HDDM are described. As illustrative examples, a support structure of the high temperature engineering test reactor (HTTR), a pressure vessel, and an internal pump of a pressure vessel are analyzed. The structure of HTTR and the pressure vessel are modeled by hexahedral solid elements whose total degrees of freedom (DOFs) are about 1.3 millions (M) and 3 M, respectively. The internal pump is modeled by quadratic tetrahedral elements whose total DOFs are about 2 M. The elastic–plastic analysis of a simple cube with 10 M DOFs is also carried out. Both the conjugate gradient method for solving the linear equations and the Newton–Raphson method for solving nonlinear problems successfully converge.

Large scale drop impact analysis of mobile phone using ADVC on Blue Gene/L

Existing commercial finite element analysis (FEA) codes do not exhibit the performance necessary for large scale analysis on parallel computer systems. In this paper, we demonstrate the performance characteristics of a commercial parallel structural analysis code, ADVC, on Blue Gene/L (BG/L). The numerical algorithm of ADVC is described, tuned, and optimized on BG/L, and then a large scale drop impact analysis of a mobile phone is performed. The model of the mobile phone is a nearly-full assembly that includes inner structures. The size of the model we have analyzed has 47 million nodal points and 142 million DOFs. This does not seem exceptionally large, but the dynamic impact analysis of a product model, with the contact condition on the entire surface of the outer case under this size, cannot be handled by other CAE systems. Our analysis is an unprecedented attempt in the electronics industry. It took only half a day, 12.1 hours, for the analysis of about 2.4 milliseconds. The floating point operation performance obtained has been 538 GFLOPS on 4096 node of BG/L.

PARALLEL THERMOELASTICITY OPTIMIZATION OF 3-D SERPENTINE COOLING PASSAGES IN TURBINE BLADES

An automatic design algorithm for parametric shape optimization of three-dimensional cooling passages inside axial gas turbine blades has been developed. Smooth serpentine passage configurations were considered. The geometry of the blade and the internal serpentine cooling passages were parameterized using surface patch analytic formulation, which provides very high degree of flexibility, second order smoothness and a minimum number of parameters. The design variable set defines the geometry of the turbine blade coolant passage including blade wall thickness distribution and blade internal strut configurations. A parallel three-dimensional thermoelasticity finite element analysis (FEA) code from the ADVENTURE project at the University of Tokyo was used to perform automatic thermal and stress analysis of different blade configurations. The same code can also analyze nonlinear (large/plastic deformation) thermoelasticity problems for complex 3-D configurations. Convective boundary conditions were used for the heat conduction analysis to approximate the presence of internal and external fluid flow. The objective of the optimization was to make stresses throughout the blade as uniform as possible. Constraints were that the maximum temperature and stress at any point in the blade were less than the maximum allowable values. A robust semi-stochastic constrained optimizer and a parallel genetic algorithm were used to solve this problem while running on an inexpensive distributed memory parallel computer.Copyright

Pareto‐based continuous evolutionary algorithms for multiobjective optimization

In an attempt to solve multiobjective optimization problems, many traditional methods scalarize an objective vector into a single objective by a weight vector. In these cases, the obtained solution is highly sensitive to the weight vector used in the scalarization process and demands a user to have knowledge about the underlying problem. Moreover, in solving multiobjective problems, designers may be interested in a set of Pareto‐optimal points, instead of a single point. In this paper, Pareto‐based Continuous Evolutionary Algorithms for Multiobjective Optimization problems having continuous search space are introduced. These algorithms are based on Continuous Evolutionary Algorithms, which were developed by the authors to solve single‐objective optimization problems with a continuous function and continuous search space efficiently. For multiobjective optimization, a progressive reproduction operator and a niche‐formation method for fitness sharing and a storing process for elitism are implemented in the algorithm. The operator and the niche formulation allow the solution set to be distributed widely over the Pareto‐optimal tradeoff surface. Finally, the validity of this method has been demonstrated through some numerical examples.