Design optimizati On: part 2 Guest editor: J S Gero In the previous issue we presented six papers on design optimization, which were all concerned with optimizing a single design objective, although the techniques which were being used varied widely. In this issue we conclude single objective design optimization with the paper by Himmelblau which provides a distillation of experience gained in applying certain nonlinear programming optimization techniques to design problems. Many designs cannot be described as having a single goal. Often we hear the client say to the designer ' . . . it should have minimum capital cost and minimum running cost'. Such a statement requires two objectives which are generally in conflict. Most designs are required to satisfy many objectives and the approaches used for single objective problems are inapplicable. Even the notion of an 'optimum' design needs to be discarded. (It is common to use the terms multi-objective and multi-criteria interchangeably.) With a single criterion to be optimized it is conceptually simple to determine the best design - it is the one which performs the best in that criterion. It can be said to dominate all other designs. With two or more criteria a statement such as 'which is the best design' may not be meaningful. Take the situation when we wish to compare two designsA and B in terms of their performances in two criteria. A may perform better than B in the first criterion but worse than B in the second criterion. How is it possible to say that one design is better than another in these circumstances? This is the dilemma of multi-criteria design optimization. The last three papers in this series address this issue in very different ways. Russell and Arlani look for a single objective which describes the goals of the
volume 13 number 6 november 1981
design reasonably adequately. Dekhtyarenko is concerned with the problem of how do you weight the various objectives so they can be added together to produce an equivalent single objective. Methods based on such approaches are common and are called preference methods. Radford, inter a/ia, is concerned with a method which does not require the articulation of preferences between objectives; such a non-preference method produces a unique set of designs, called the Pareto set, which dominate all other designs in a multi-criteria space in a manner similar to the way in which the best design dominates all others in a single criterion design. As more suitable techniques, both preference and non-preference, are developed for multi-criteria design problems we can expect to see a burgeoning of applications of design optimization in CAD. This will produce changes not only in the design process but also in the design resultant: changes which are much more significant than those being wrought by changes in those areas concerned with design documentation productivity.
BIBLIOGRAPHY Gero, J S and Radford, A D 'The "design" in computeraided design' Comput. in Civil Engineering ASCE, New York (1981) pp 876-890
Gero, J S, Radford, A D and Murthy, N S 'What if?, exploring the consequences of design and performance decisions in computer-aided design' Working Paper, Department of Architectural Science, University of Sydney, Australia (1981 )