EFFICIENT DESIGN USING FUZZY-LOGIC BASED REGRESSION-MODELS

With ever decreasing design cycles, it is important for designers to h ave techniques they can use to quickly and efficiently model new desig ns. From these models, package performance can be estimated and electr ical, thermal, and mechanical considerations can be balanced. In this paper, we present a method of quickly investigating new design concept s based on knowledge of previously studied designs and knowledge of th e differences between the new and old designs. This approach is useful when the difference between designs is simple and can be accurately m odeled with fewer data, The use of less data equates to a savings of t ime and money. In the case studies presented, we establish two ''base' ' models using two data sets of 40 points each, then we establish two additional models of similar processes using only five and seven point s each. Here, the initial (base) designs and the design differences ar e modeled with fuzzy logic based regression models. Such fuzzy logic b ased regression models can be based on numerically or empirically obta ined data and/or qualitative knowledge of the system to be modeled. On ce established, these models have the advantage of offering very fast response times uncharacteristic of experimentation, prototyping, and n umerical methods such as finite element, finite difference, or boundar y element modeling.