Design Window Search Based on AI Technique

抄録

The design process of a new reactor consists of broad technical areas. The work of each area may be carried out independently to achieve design target or to satisfy design criteria. The decisions made in one area will affect the design basis in most other area. The design process is thus iteratively performed. To satisfy all design criteria, many parametric survey calculations should be carried out in general. In any case it would be difficult to find an optimal set of design parameters. Such a situation in design problems would be fairly different from that in anomaly diagnosis problems with respect to an application of the AI technique. To reduce the number of iterations at the stage of the design study, it is useful to draw a design window in which the design is feasible. Any parameter set inside a design window area can satisfy design criteria or goals. Design windows obtained for various combinations of design parameters in each area help designers with the decision of design parameters. Even in this case, a considerable amount of calculations must be performed by varying parameter values.
At the present work, we intended to search automatically a design window with use of AI based techniques in order to save cost and time. One of the authors has proposed the design window search algorithm for a reactor structural problem and shown its validity⁽¹⁾. Since there are many design parameters, a design window is generally a sub-space of multi-dimensional one. As the first step, we investigated a procedure to search a window in two-dimensional space, that is, for two parameters. The search procedure and an example applied to the thermal hydraulic design are described below.