Constraint programming languages: their specification and generation
Constraint programming languages: their specification and generation
Network-based heuristics for constraint-satisfaction problems
Artificial Intelligence
Hi-index | 0.00 |
Conventional methods for the parametric design of engineering structures rely on the iterative re-use of analysis programs in order to converge on a satisfactory solution. Since finite element and other analysis programs require considerable computer resources, this research proposes a general method to minimize their use, by utilizing constraint-based reasoning to carry out redesign. A problem-solver, consisting of constraint networks which express basic relationships between individual design parameters and variables, is attached to the analysis programs. Once an initial design description has been set out using the conventional analysis programs, the networks can then reason about required adjustments in order to find a consistent set of parameter values. We describe how global constraints representing standard design behavioral equations are decomposed to form binary constraint networks. The networks use approximate reasoning to determine dependencies between key parameters, and after an adjustment has been made, use exact relationship information to update only those parts of the design description that are affected by the adjustment. We illustrate the ideas by taking as an example the design of a continuous prestressed concrete beam.