| Today fully exploiting knowledge has become a primary opportunity for the manufacturing industry to reduce cycle time, production cost, and achieve competitive advantages. With the explosive growth of our capabilities to both generate and collect data, data mining (or knowledge discovery in databases) has emerged providing a scalable and automated way of knowledge creation directly from historical data. The goal of our research is to help our industrial partners extract explicit, understandable and useful domain knowledge from data generated by complex design and manufacturing processes, and furthermore make effective knowledge management.; This dissertation presents a novel evolutionary approach for learning accurate, compact, and noise-tolerant classification rules based on our enhanced version of gene expression programming (GEP), a new hybrid of genetic algorithms (GAs) and genetic programming (GP). Compared with traditional tree-based GP, our GEP approach is more efficient and the solutions are much more simple and therefore easier to understand. Unlike most traditional data mining algorithms, GEP exhibits great capability and flexibility, which can discover higher-order relationships in the form of any logical and mathematical combination of attributes and express them mathematically.; We also propose an integrated framework that exploits synergies between knowledge management and data mining in order to facilitate and automate the full lifecycle of knowledge creation, representation, maintenance and reuse. Hybrid data mining approaches are seamlessly embedded into the knowledge management process, such that robust concept learning capabilities can be achieved for a variety of difficult manufacturing tasks. In addition, a two-round meta-learning method is proposed to automatically derive meta-knowledge for model reuse, which specifies under what circumstances each individual data mining technique is applicable. Moreover, a component called Knowledge Container is presented, which not only stores different types of domain knowledge, but also maintains the knowledge source and meta-knowledge for over-time usage of knowledge. |
