Design of an Expert System for Emergency Response to a Chemical Spill. 1. Domain Definition and Knowledge Acquisition

Abstract

This paper describes the knowledge acquisition stage in the development of an expert system designed to assist in planning the most appropriate response to chemical spills. A knowledge orientation analysis has been applied to the problem domain and a conceptual decision-making hierarchy has been constructed.Five primitive phases of action have been identified from this conceptual model, and these phases are followed in the subsequent development of the prototype. The knowledge base that was developed in this work consists of a database of factual information and a rulebase of heuristic knowledge. Using a subject-oriented structure, the database of information can effectively handle the large amount of literature information necessary to be considered in planning the response to a spill. The heuristic knowledge is represented as a set of production rules that help in the interpretation of the factual information. The knowledge acquisition process is referred to as a bottleneck in the development of expert system applications. For this prototype the knowledge acquisition step is carried out using an innovative knowledge encoding scheme, known as the knowledge domain matrix (KDM). A fully functional knowledge base has been developed.+ Keyword: knowledge acquisition, chemical spill, environmental emergency, knowledge base, production rules, inference engine, knowledge domain matrix, ERexpert.

Expert Systems and Analytical Chemistry: Recent Progress in the ACexpert Project

“…Expert systems have also been applied to a number of other areas. Applications that allow evaluation of physico-chemical properties, 123 characterization of polymers, 124 converting chemical formulas to names, 125 and handling emergency chemical spill accidents 126,136 have been described. Some standard procedures in the chemical decision-making stream can be taken over by expert systems, improving the performance and productivity in laboratories as well as increasing the knowledge sharing probability.…”

“…Unlike factual knowledge, heuristic knowledge does not have a single well defined value and is hard to express and document precisely. 136 One way, and probably the only effective way, to become familiar with heuristic knowledge that is used in a specific domain of expertise is through application of the knowledge in problem solving processes. Experts are experts because they know both the factual and heuristic knowledge that defines their domains.…”

“…Our own results and the reports from other research groups have suggested that the decision-tree structure adopted by most expert system shells may be ineffective for knowledge encoding. 86,110,136 Not only are the logical relations presented in a tree structure rigid so that any changes made to the tree may upset the total relationship but also the connections in a decision-tree are obscure, so that a knowledge engineer may not be able to understand the structure easily. The only feasible way to update a knowledge base developed in this way may be through a major revision from root to tip.…”

Expert System for Emergency Response Design of an Expert System for Emergency Response to a Chemical Spill. 2. ERexpert Module Design and Development

“…The Factbase comprises a number of subject-oriented tables in which factual information is stored as either text or numbers. 20 During execution, the heuristic module containing "IF condition THEN conclusion" rules can be used to offer the user advice based on the facts provided, while a search of the Factual database provides information specific to the case that can supplement the conclusion(s) reached during an inference process. 3.2.…”

“…In reality, however, the decision-making process may often require multiple inferential processes to be performed at different levels, and each of these levels may require the use of previous conclusions, together with additional observations as new conditions, to carry the reasoning process on further. 20 The diagnostic procedures necessary to solve a fault in an hyphenated instrument, such as GC MS, requires, at a minimum, a two-step inferential processes to be performed in tandem. In terms of the KDM, this multistep inference process cannot be represented by a single twodimensional matrix.…”