DECADE: A HYBRID KNOWLEDGE-BASED SYSTEM FOR CATALYST SELECTION (FISCHER-TROPSCH REACTION).

摘要

DECADE (Design Expert for CAtalyst DEvelopment) is a prototype knowledge-based system for catalyst selection. The objective of DECADE's development has been to investigate and evaluate the potential of knowledge-based systems technology applied to the solution of chemical engineering problems. DECADE's particular application problem consists of prescribing a set of catalytic materials and operating conditions that have an acceptable probability of being appropriate for a target reaction (to be accomplished in a single reactor). The class of reactions for which DECADE has specific knowledge is carbon monoxide hydrogenation.; Given DECADE's architecture and implementation, it can illustrate the integration of different paradigms along some of the several dimensions of knowledge-based systems applications: (1) In terms of knowledge representation: DECADE uses three different representation mechanisms, each one appropriate to its function: (a) rule-based for the overall control, interaction among parts, and inferential steps. (b) frame-based for the description of concepts (their properties and relations). (c) functions and numerical algorithms for the calculation-intensive steps. (2) In terms of problem-solving methods: The selection of catalytic materials requires several different kinds of problem solving methods, for example: (a) straight algorithmic calculation is used in the thermodynamic feasibility calculation section. (b) diagnosis is necessary in the reaction classification step. (c) means-ends analysis is used in the prediction of reaction mechanism steps and surface intermediates. (3) In terms of levels of knowledge abstraction: (a) Reaction level. (b) Molecular level. (c) Catalyst surface level. All these properties are achieved through the use of different languages (OPS5, SRL1.5, Franz Lisp) brought together in a blackboard model architecture. The size of the system is roughly 300 production rules, 325 frames, and 200 functions.; The most interesting results are produced at the deeper level of abstraction (which relies only on medium to deep level knowledge). Catalysts are proposed at this level using a generate and test procedure with a priori and dynamically generated constraints. Explanation of the results is available for any material that was taken into consideration (whether recommended or not). Taking into account the limited size of the knowledge base, the results for this level of abstraction are acceptable to good.