A knowledge system for automatic finite element mesh generation: AMEKS

Finite element analysis is widely employed by the engineering community for the numerical analysis of complex structures. It involves the partitioning of an object into small elements for which applicable physical responses can be readily formulated. The generation of this partition, or mesh, is a tedious and often difficult part of the analysis. This paper presents a knowledge-based system, AMEKS, that, in conjunction with existing software, automatically meshes objects using the transfinite mapping technique. This technique requires tha analyst to decompose complex objects into primitive shapes and prepare each primitive for algorithmic meshing based on problem geometry and mesh requirements. AMEKS now performs these tasks. Several characteristics of this problem indicate the need for artificial intelligence techniques. Decomposition is non-algorithmic and expertise intensive. The number of possible decompositions is usually too large for exhaustive search. Locally acceptable decompositions can prove to be globally unacceptable. The concept of ''primitive'' is ill-defined, and once a region is classified as primitive, its acceptability for meshing is case-by-case dependent on geometry, physics and the available alternatives. AMEKS currently accepts two-dimensional straight-sided objects. It uses frames to represent the geometry, rules to classify regions and evaluate region quality, heuristic search control to prune the large more » search space, and assumption-based truth maintenance to simultaneously examine multiple contexts during the decomposition. « less