Generative and evolutionary design exploration

Expert designers, both architects and engineers, typically display a strategy of exploring design alternatives, albeit a relatively small number. Expert architects’ strategy in problem solving can be denoted breadth first, depth next, in comparison to novices, who typically display less breadth of exploration (Akin, 2001). Although engineers’ strategy is markedly different, design alternatives play a role that is as important, if not more important, than for artchitects (Akin, 2009). Where designers typically consider a very small number of alternatives in their work, this can be explained by cognitive limits, opening the door for computational support of design exploration. In particular, Woodbury and Burrow (2006) argued that exploration is a compelling model for designer action and that designers benefit from tools that amplify their abilities to represent goals and problems spaces, and to search for designs. Generative and evolutionary methods have proven to be strong catalysts for design exploration, and design optimization has served as a means to assist in this exploration. Recently there is a marked move toward using optimization to aid exploration. Optimization is rarely intended to yield an optimal solution per se, instead assisting in gaining insight in the solution space, thereby reducing the size of the solution space for exploration, possibly focusing attention toward the Pareto boundary. Even at the Pareto boundary, there are a large number of solutions worthy of further exploration. Exploration and optimization together lead to a better understanding of the complexities of design issues and help designers in their decision-making process, especially with multiple-objectives problems, which is a nature of many design problems. As such, the focus of attention in generative and evolutionary design is shifting from the techniques themselves, and their direct application, to the way we are using these techniques to assist and improve the design and engineering process. We might frame generative and evolutionary design from the point of view of a “conversation” in the sense of Donald Schön (1983); this is nothing uncommon for generative design, though it is for optimization. This type of conversation is between the designer (or design team) and the computer, and is digitally enhanced. As such, the aim is less on optimization per se and more on exploration: the results from optimization are about changing one’s way of thinking more than choosing a single design and then realizing it. We can then ask the question of how these types of conversation can unfold. How do they start and where do they end? What to do with thousands of similar solutions? The 11 papers in this Special Issue all address generative and evolutionary design exploration and contribute to the discussion of the interaction between design exploration and evolutionary design optimization. Julian Eichhoff and Dieter Roller start off with “A Survey on Automating Configuration and Parameterization in Evolutionary Design Exploration.” Focusing specifically on engineering design, they comprehensively review evolutionary design optimization approaches based on genetic algorithms (and genetic programming) addressing different design phases. Reformulating design problems as multiple-objective design optimization problems commonly reduces to parameterization, defining constants, and configuration, defining design variables, objective functions, and constraint functions. Methods from the fields of machine learning and natural language processing are reviewed to support these parameterization and configuration processes. Herm Hofmeyer and Juan Manuel Davila Delgado, in “Coevolutionary and Genetic Algorithm Based Building Spatial and Structural Design,” compare the use of a genetic algorithm with a coevolutionary method to collaboratively develop and optimize building spatial and structural designs. The genetic algorithm uses a finite element analysis method for evaluation of design alternatives. The coevolutionary method applies deterministic procedures to cyclically evaluate and improve the structural design via a finite element method and topology optimization, and adjust the spatial design according to the improved structural design and the initial spatial requirements. Both methods provide optimized building designs; however, the coevolutionary method yields Reprint requests to: Rudi Stouffs, Department of Building Technology, Faculty of Architecture, Delft University of Technology Postbus 5043, 2600 GA Delft, The Netherlands E-mail: r.m.f.stouffs@tudelft.nl Artificial Intelligence for Engineering Design, Analysis and Manufacturing (2015), 29, 329–331. # Cambridge University Press 2015 0890-0604/15 doi:10.1017/S0890060415000360