抄録
A huge amount of calculation is needed for optimizing design of actual structures. Various studies on parallel and distributed calculation have been performed in relation to a collection of computing resources distributed on a local area network. Recently researches on a highly efficient wide area computing system called the "Grid" have appeared. In order to realize design optimization of artifacts on the Grid, the following requirements have to be satisfied : (1) Computation with little dependency can be performed asynchronously, and (2) An optimal solution for required accuracy can be obtained robustly and promptly. The present study describes the development of a robust and efficient system for optimization of structural design of artifacts on the Grid. To develop the system, both the requirements are considered simultaneously. To satisfy the requirement (1), evaluation calculations such as finite element structural analyses are asynchronously performed on the Grid. Each FE analysis is solved using an Open Source parallel finite element system named ADVENTURE, developed by the authors. To satisfy the requirement (2), two kinds of efficient optimization algorithms based on real-coded genetic algorithms are developed. Both methods devise generation of new sample points. The first algorithm employs approximate quadratic programming problem, while the second one employs smoothing spline. In addition to these algorithmic works, robust processing mechanism is implemented to realize a whole computation on the Grid without interrupting. In the study, two kinds of Grid environments are employed. The one is simply a group of PC clusters distributed in remote sites. The other is one of Japanese national Grid projects named ITBL (IT-based Laboratory). Practical performances of the present system are demonstrated through some examples.
