RECURSIVE ANNEALING - A COMPUTATIONAL MODEL FOR MACHINE-DESIGN

We propose a model for optimally directed conceptual design of machine s in which the transformation of function to form occurs iteratively a long an abstraction continuum. An algorithm called FFREADA is introduc ed as a computational implementation of the model. FFREADA is a gramma r-based optimizing design algorithm that uses recursive simulated anne aling to generate optimally directed designs. During FFREADA's design process, the mapping of function to form is accomplished using an abst raction grammar production system and a predefined library of function and form entities. FFREADA also has a random design generation mode t hat can be used to record data to characterize the space of design sol utions. FFREADA is demonstrated by designing an idealized power supply using a variety of performance objectives. Results show the algorithm able to explore and record information about a tractably infinite des ign space before converging to the optimal design.