Qualitative modeling and simulation methods provide the ability to express incomplete knowledge of the structure and initial state of a mechanism, and to derive useful predictions about its possible behaviors. This capability is of major significance for model.based reasoning in diagnosis, design, and decision.making. The proposed work builds on the investigator's prior development of the QSIM representation and algorithm for qualitative simulation. In the model.simulation area, the new work focuses on these problems: (1) the interpretation of quantitative data in terms of qualitative predictions; (2) representing qualitative behaviors as trajectories in a qualitative phase space, allowing application of more powerful mathematical methods; and (3) hierarchical structuring methods, such as abstraction of mechanisms by relative time.scale, to deal with complex mechanisms. In the model.building area, the investigation concerns automatic methods for building qualitative models, focusing on two major issues: (1) the selection, from potentially vast library of tangentially relevant descriptions, of a small set of descriptions to incorporate into a useful model; and (2) reasoning with the limits of the model, to know when it is useful, and combining conclusions from multiple models.
