An expert system for developing and evaluating photochemical reaction mechanisms

Four problems arising in the evaluation of photochemical mechanisms are: (1) lack of adequate attention to the detail of each individual chamber experiment, (2) lack of adequate analysis of goodness-of-fit of mechanism against data, (3) failing to present simulation results at different levels of detail, and (4) failing to provide a complete ad traceable inference chain linking test results to the conclusion. An expert system approach was used to address these problems in a comprehensive manner. This approach promises to make the mechanism evaluation process systematic, rigorous, and explicit. Three sources of knowledge were used in formulating such an approach: photochemical modeling, expert system techniques, and theory of evidence.; The Dempster-Shafer theory of evidence was used to propagate the uncertainty inherent in the model evaluation process. This theory forms the basis of a generalizable scheme to determine the goodness-of-fit of simulation results against data. A relatively new extension of the Dempster-Shafer theory known as quasi-compatibility relations was also used to construct a rule base which contains the arguments for the acceptability of a mechanism. One big advantage of using the quasi-compatibility relations method is the ease with which beliefs can be propagated up a set of rules connected in a hierarchy and the flexibility with which beliefs can be attached to the propositions triggering the rules.; A practical scheme was used to reduce the computation cost of inference systems employing the Dempster-Shafer theory. This scheme reasons about a reduced power set of the frame of discernment and makes the computation cost increases only linearly with the increasing size of the frame of discernment.; An expert system called ASKME for Atmospheric Simulation Kinetics Model Evaluator was implemented using the approach. The system was tested systematically and found to be a reliable tool to assist in the multi-variable judgmental task of determining the goodness-of-fit of a model and in reaching a conclusion about the acceptability of the model with supports at different levels of detail.