Model-based problem-solving and semantic control of the ATMS (Assumption-Based Truth Maintenance System)

Extending formal techniques to solve large, real problems is difficult because of the size and complexity of the state spaces. As size increases, solution techniques having the desired formal properties are often useless or impractical because of the frame problem, interacting subgoals, and combinatorial explosions. This paper presents a Model-Based Problem Solver (MPS) that efficiently builds solutions of the desired quality for a class of large, real problems. The MPS semantically evaluates partially constructed direct models to avoid the frame problem to determine the correct ordering of operations, and to control the creation of Assumption-Based Truth Maintenance System (ATMS) context space for generating finite element meshes. Automatic quadrilateral finite element mesh generation is a long-sought engineering analysis capability. The Automated MEshing Knowledge System (AMEKS) automatically decomposes 2-D regions, with boundaries composed of straight and curved lines, into primitive shapes that can then be meshed algorithmically. The MPS does this in linear time, even though the state space grows exponentially with the number of solution steps. In one example of a real part, an exhaustive search would require more than 54,000 ATMS contexts. AMEKS constructs and recognizes a useful decomposition using only 26 contexts. 11 refs., 6 figs.