Publications

An artificially intelligent approach to conceptual design, from initial problem specification to the generation of a good concept, is presented and applied to thin-walled fibre reinforced composite panels for aerospace structures. Conceptual design is a creative and multi-disciplinary process for which numerical computer aided design tools are not sufficient. Specifications comprise both quantitative and qualitative objectives and limitations. Due to the diversity in application of a design system, specifications, and in experience and preference of designers, there is not a predefined procedure for this design process. However, 3 consecutive design tasks are distinguished.

The first task is to propagate the specifications to limit the design space. Constraint-based reasoning is used to assemble prototype concepts for the specifications on qualitative behavioural relations. The second task is to find the region of optimal and practical designs. Case-based reasoning is used to retrieve exemplar concepts for the prototypes. The third task is to heuristically optimise retrieved concepts for the current specifications. Rule-based reasoning within the concept of Expert Governed Numerical Optimisation (EGNOP) is used to modify the concepts according to explicit heuristic rules. Once a good initial concept is obtained and quantified that satisfies all the qualitative specifications, existing numerical design tools can be applied.

The approach is developed as a supporting tool for the designer, that performs more routine-like reasoning and suggests decisions to the designer. The approach enables multiple knowledge representation techniques and multiple inferencing techniques to more adequately model the domain knowledge and design process.