Adapting conceptual clustering for preliminary structural design

The use of conceptual clustering for knowledge acquisition can facilitate the development of a knowledge base in a domain where little formalized knowledge is really available. The limitations of current conceptual clustering techniques include a lack of accommodating varied attributes across training examples and a lack of learning associations among attributes within a cluster. A methodology is presented in which conceptual clustering is adapted to satisfy these limitations. A preliminary conceptual clustering technique is described where training examples are grouped according to similarity of attributes, rather than similarity or utility of similar values. Conceptual clustering is augmented by numerical methods for linear regression analysis and probabilistic approaches to pattern identification. The methodology is illustrated through its application to learning concepts for the preliminary design of bridges.