4.1 KNOWLEDGE-BASED EVALUATION FOR DESIGN
Knowledge engineering proposes that standards and proven heuristics within a given domain of expertise be encoded into representations that are computationally manipulatable. Knowledge engineering is based on symbolic computation as opposed to numeric computation.
Knowledge-based systems of were once commonly referred to as expert systems. But the true experiential nature of the expert, it has been realized, can not easily be approached with the offerings of current A.I. technology. Hence, the claims of matching the abilities of an expert within a given system have been muted to claims of the system at least being based upon the knowledge of an expert.
Some expert systems specifically developed for plastic injection molding tend to provide analysis similar to that of analytical/numerical simulation, while providing good heuristic compensation for the recognized weaknesses of numerical simulation. These systems take a few key parameters such as the length, width, height, and material type to perform analysis [Bernhardt 87][Manzione 87]. As with numerical models, these results are based on simplified dimensioning of a part, the results may provide only first order insight into potential problems.
The Plastic Injection Molding Expert System (PIMES) advisor is responsible for DFM's critical evaluation of a part. The approach of the PIMES agent is to utilize a knowledge base containing rules that formally represent heuristics and design standards used by experts in the design and manufacturing domain. These heuristics, or rules, rely on an object's form features to determine whether the feature parameters violate any design or manufacturing constraints. A form feature as a part of an object that is physically different from the rest of the object and performs certain functions [Dixon 88][Frederick et al 83].
Rules are formatted and stored as productions. These productions primarily inference upon symbolic objects which represent the features. While these rules generally center on detecting constraint or specification violations, they avoid encoding specific facts. The rules are generic enough to apply to certain features in that they essentially devoid of actual values or facts. Although the actual facts are derived from the part design, values required for expert system reasoning are obtained from intermediary databases. PIMES avails itself of additional production information, such as material specifications and optic finishing standards, which are stored in a database.
The PIMES system's knowledge, as frames and rules, are store in several knowledge-bases. The rules and frames themselves are developed within Nexpert Object, a commercially available expert system shell. Nexpert utilizes frame representations, with an object-oriented paradigms of pre-defined classes, properties, inheritance methods, and triggers for self-contained entities. Its production-based operators are involved in configured forward and backwards inferencing. And extensive object set processing capabilities characteristic of Fourth Generation Languages (4GLs) are used. PIMES makes use of all these capabilities, although it limits itself largely to backwards inferencing.