Composition, Management, and Exploration of Computational Studies at Early Design Stage
Marin D. Guenov, Marco Nunez, Arturo Molina-Cristóbal, Vis Sripawadkul, Varun Datta, Atif Riaz · 2014
Exploring the design space with the aim of finding feasible, let alone optimal solutions is not trivial. Essentially, this involves finding a solution to an inverse problem. That is, a relatively small number of (presumably) known requirements and performance characteristics need to be mapped onto a much larger number (space) of unknown design parameters, subject to constraints. The inverse problem, of course, is generic and forms parts of the study of complex systems, from ecology [1,2] to engineering design [3], where the integration of an expanding set of (validated) numerical methods and models is used to investigate scenarios and to predict outcomes. The work presented here lies within this context, with a particular emphasis on computational intelligence methods and tools for interactive design space exploration. The scope is restricted (but not limited) to conceptual computational design where a complex product, for example aircraft, ship, and so on, is described by a large number of computational models related to geometry parameterization, performance, cost, and so forth. It is assumed that the computational models are black-boxes (e.g., compiled code) which contain low-fidelity code (e.g., parametric/empirical equations) and/or surrogate models. This assumption reflects the realities of the commercial world in which the content of a model is usually a closely guarded intellectual property. There are a number of challenges associated with such complex and relatively little studied computational systems: