The integration of model-based reasoning and task-specific problem-solving architectures for chemical process diagnosis.

摘要

Knowledge-based systems for chemical process diagnosis are required to produce timely and accurate results. Problem solving systems based on task specific architectures organize knowledge and inference for efficient generation of diagnostic advice. Accuracy has historically been associated with reasoning about process structure and behavior as demonstrated in model-based diagnosis. The goal of this research has been to integrate model-based reasoning into the current compiled generic task architecture used for chemical process diagnosis in order to achieve enhanced accuracy and efficiency. To characterize the design of the integrated architecture, model-based diagnosis and compiled generic task diagnosis are compared in an effort to identify what attributes contribute to the accuracy and efficiency of a diagnostic problem solving system.; The site of integration for model-based reasoning and the compiled task specific architecture occurs in a class of problem solving situations called malfunction reenactment. Diagnosis is performed by the task specific portion of the architecture, followed by model-based reasoning about process structure and process behavior in the context of the malfunction(s). This class of situations includes primary interacting malfunctions, secondary malfunction interactions, partially canceling malfunctions and corrective action planning in the context of malfunctions. A comprehensive integrated architecture is presented that applies task specific complied reasoning and model-based reasoning to the resolution of interacting malfunctions, the first two types of scenarios involving malfunction reenactment. The problem solving architecture performing integrated model-based reasoning is called Diagnostically Focused Simulation (DFS) and is composed of a collection of problem solving modules that reason about structure and qualitative behavior under the focus and guidance of a compiled problem solver. The description of each module stresses the knowledge and inference requirements as well as the impact of compiled reasoning. The complete problem solving procedure is demonstrated with interacting scenarios from the chemical process domain. DFS combines compiled and model-based reasoning in a tightly woven fashion, leveraging the strengths of each methodology in a effective and systematic manner.