|
The ability of Genetic Algorithms (GA) to identify local and global optima makes them especially suitable for solving structural system reliability problems that essentially can be reduced to optimization problems in the standardized normal probability space. To overcome the known inefficiencies of traditional Genetic Algorithms, this presentation proposes the use of a hybrid search technique that will be demonstrated to have the capability of efficiently identifying the dominant failure modes of a structural system and estimating their reliability index values.
The proposed hybrid algorithm includes a newly developed GA linkage learning process that essentially weeds out less fit individuals by tightly binding together two or more closely correlated genes allowing them to travel as one unit under the action of GA operators. In addition, the linkage process is used as an expert system to help identify fit search directions without the need to perform computationally costly exact evaluations of the fitness. The application of the proposed linkage learning process will be shown to overcome many of the limitations associated with GA such as building block disruption, inadequate exploration, spurious correlation, and many other perceived stumbling blocks to the widespread implementation of GA in structural applications.
The presentation will illustrate the high efficiency and robustness of the proposed algorithm for solving structural reliability problems using benchmark tests and by application to several realistic models of complex structural systems.
|
