Exploring spaces of system architectures using constraint-based classification and euler diagrams

Abstract

In many situations it is necessary to explore a set of system architectures, networks of elements linked by directed or undirected relations, where both elements and relations may have types assigned to them. The architectures may represent products in a portfolio or across competitors, or alternative process plans for undertaking a project. However, exploring spaces of system architectures is difficult because they are qualitative structures, and thus are not easy to compare directly. As a proxy for direct comparison, they may be compared using metrics to assign numerical values to them (Lindemann et al., 2008), but such metrics may be difficult to interpret in concrete terms or may require potentially unavailable information (Wyatt et al., 2009a). Alternatively, architectures may be clustered by some measure of "similarity" (Langdon and Chakrabarti, 2001), but strict hierarchical clustering may not recognise the complex similarity relationships that may exist between different architectures. A third alternative is the use of methods such as ΔDSMs (de Weck, 2007) to highlight the differences between two architectures; however, these methods do not allow larger sets to be compared directly. Therefore, a need remains for methods to compare structural characteristics of architectures in larger datasets.