Using Modeling Knowledge to Guide Design Space Search

Virtual Design, Problem Framing, and Innovation: An Empirical Study in the Automotive Industry

“…We observed that the problems related to designing a hybrid propulsion system were similar for both companies, but were framed in different ways. We used the concepts of design space and degree of granularization as problem framing measures (Gelsey et al., 1998; Smith and MacLean, 2007; Stankiewicz, 2000), and to make sense of the differences between projects: Figure 1 shows that the designs for the new hybrid propulsion systems in Projects T and M were based on combinations of different sets of components and subcomponents. Project M's design involved combination of six large components; Project T's involved 15 components.…”

“…In this paper, we use the degree of granularization of the design space as the focal measure for problem framing. Granularization is the decomposition of a given problem into coarse or fine sub‐problems (Gelsey et al., 1998; Smith and MacLean, 2007; Stankiewicz, 2000). This variable has been shown to be a reliable lens through which to study problem framing, within a vast literature spanning technology management (see, e.g.…”

“…Stankiewicz, 2000), human–computer interaction (see, e.g. Smith and MacLean, 2007), and artificial intelligence (Gelsey et al., 1998). Granularization is a useful lens because it provides an immediate link to new product development conceived of ‘combinatory of elementary cognitive and physical components’ (Dosi and Grazzi, 2006, p. 179).…”

“…Our research identifies three organizational/strategic variables that affect problem framing and, in particular, the degree of granularization of the design space, i.e. whether the ‘problem’ is decomposed into many small or a few large sub‐problems (Gelsey et al., 1998; Smith and MacLean, 2007; Stankiewicz, 2000).…”

Learning symbolic formulations in design: Syntax, semantics, and knowledge reification

“…The development of artificial intelligence (AI) algorithms to automate design problem reformulation tasks is an enduring challenge in design automation. Existing methods either require dependence upon high levels of embedded knowledge engineering in the form of rules, heuristics, grammars, or domain/taskspecific procedures (e.g., Ellman et al, 1998;Gelsey et al, 1998;Medland & Mullineux, 2000;Campbell et al, 2003) or require a large database of training cases (e.g., Duffy & Kerr, 1993;Schwabacher et al, 1998). It would be useful to develop a method characterized by the following desirable features: a knowledge-lean method that does not need any significant design domain or task knowledge to be embedded into the system; a training-lean method that can extract design knowledge over one or very few cases; and a simple and computationally efficient method applicable over different design domains, representational forms (analytical, nonanalytical, etc.…”