| Today's business environment is one of cheap computer storage and widespread data collection. The sheer volume, heterogeneity, and distributed nature of stored data, coupled with obsolete legacy software systems, has hindered the efficient use of this precious information. Data mining, a field related to machine learning, finds patterns and relationships in such large, multi-dimensional databases.; At present, product designers do not have easy access to product life cycle information and other critical feedback contained in stored data, resulting in costly design iterations and increased time-to-market. In variant design, where designers are reusing components and subassemblies, historical data related to earlier iterations of the product contain crucial knowledge that can lower the total cost of product design.; This research proposes a data mining approach to incorporating heterogeneous and distributed product information, within a system framework that allows the designer of a product variant to access both mined knowledge and original documentation. Previous studies in knowledge-based methods largely ignored how the knowledge is acquired, but recognized the difficulty of acquisition and maintenance of domain information. Data mining methods are an excellent solution for this bottleneck. However, existing data mining methods may not be suitable for engineering data, requiring fundamental research into new algorithms that can handle these data structures.; In addition to the systems architecture contribution of this research, a major portion is devoted to a clustering algorithm for bills of material (BOMB), which are represented as unordered trees. Once similar BOMs have been clustered, a unification operation builds a generic bill of materials (GBOM) for each cluster. The GBOM is then mined for manufacturing and design rules and represented as a constrained XML (cXML) file. Manufacturing and design constraints are checked to make sure they are satisfied, and the file is then used to search for similar parts, components, and designs. Case studies using industrial data evaluate the effectiveness and efficiency of this approach.; Also addressed as part of this research is the issue of adaptive data mining, management of mined knowledge, and representation of design rules. Finally, recommendations are made for further work in variant design support. |
