Utilising Abstract Matching to Preserve the Nature of Heuristics in Design Optimisation

Design space exploration, the generation of alternate designs to identify working designs with varying system properties, has the potential to provide a basis for the optimisation of computer-based system architectures. To utilise design space exploration for this purpose requires that an effective mechanism exist for the storage and application of potential design changes. Heuristics have shown some promise in this area due to their ability to capture expert design knowledge and their flexibility across multiple domains. Heuristics are also especially attractive as change descriptions as they can capture changes that operate across a large spectrum of change detail, from the very detailed to the very abstract. Heuristics are, however, at their most powerful, and their most useful, when they are specified in an abstract manner. This presents a challenge in the formal application of heuristics in capturing design knowledge. Formally describing heuristics inclines their specification of change to be in a more detailed, more concrete, state than an abstract one. This occurs because architectural models tend be both domain specific and are often described at a more concrete level than the level at which the heuristic is described. This has the potential to greatly reduce the effectiveness of heuristics. In this paper we propose that by providing an abstract match method heuristics may be specified in an abstract manner and still be applied to a detailed formal model, thereby eliminating this problem.