Abstract
In an industrial design, it is necessary for most products to satisfy a cost, product performance, weight, a design, and other functional requirements at a time; how to get balance is important while a product satisfies their functional requirements. Recently, with the development of the computer simulation technology, the simulation is performed before demonstration experiment, and a method to understand a physical phenomenon beforehand in a certain level of range is used. However, a flood of calculation time and cost run design itself depending on a problem when the detailed model of the product is designed by simulation. Therefore it is impossible practically to optimize for a detailed model. Therefore, an experiment and the simulation are performed with experimental design, and a response surface method predicting space of function is suggested as one of the effective plan by come out of simulation. The making of the response surface model has various methods, a study of the sequential approximate optimization technique that used the Convolution Radial Basis Function (CRBF) network that can be good approximate nonlinear function. Sequential approximate optimization technique is a method in search of a precision solution by repeated to approximate object function by a number of sample points and to predict optimized solution on the basis of object function optimization. We must regard about below when approximate optimization is actually operated. The first is that how structure an approximate function. The study can to approximate that have the advantage that tackle to benchmark problem without base function and very simple and high-speed learning by approximate of object function by using CRBF network. The second is that how sampling method sample point to have approximate function with satisfactory accuracy and lowest possible sample points. The study, add in a sample points of around 1/3 in decreasing order of the error with the set targeted value. The third has a point that regard is necessary for the manipulation of constrained condition, this is necessary for this to give a suitable condition for analysis and evaluation of the benchmark problem. Presently, this sequential approximate that used the CRBF network practical realization of industrial design that used this optimization technique. In this study, we have optimal industrial design than ever before in setting the criteria by the sequential approximate optimization technique that used the CRBF network to solve a varied benchmark problem.
