| Design and manufacturing knowledge, including proprietary practices and trade secrets, is critical to the success of an industrial organization. A good Knowledge-Based System (KBS) can easily acquire, organize and update design and manufacturing knowledge and, as a result, can help reduce product design time and potentially cut the cost of creating new products. However, the development of a KBS carries high cost since it requires specialized knowledge engineers to interact with domain experts for a sufficiently long period of time; this is widely known as the knowledge acquisition bottleneck . In addition, the construction of a KBS has a high risk since the developed system might have only limited capabilities (biased towards the knowledge engineer's understanding or the expert's experience) and may quickly become obsolete (when new equipment or process is adopted). More effective and efficient knowledge acquisition methods should be developed and applied for the construction of a KBS.; In this research, barriers in a KBS are explored and a methodology of integrated knowledge acquisition is proposed. This methodology integrates neural networks, decision-tree learning, and fuzzy logic, to automatically acquire knowledge in terms of highly intelligible and concise rules from empirical/experimental data. Using the proposed data discretization, data conversion, rule extraction, and rule simplification algorithms, this methodology can directly extract high-level IF-THEN rules from the database, thus allowing knowledge engineers to independently acquire rule-type knowledge and become “pseudo-experts.” The proposed methodology is expected to greatly shorten the system development cycle by allowing a more efficient interaction of knowledge engineers with domain experts.; The efficiency of the knowledge acquisition approach was demonstrated by its application to a real-world manufacturing problem—process planning for drop hammer forming process. |
