Hybrid Data Mining/Genetic Shredding Algorithm for Reliability Assessment of Structural Systems

Abstract

Recent studies have successfully introduced genetic algorithms (GA) to identify the important failure modes of complex structures and quantify their contributions to the reduction of the reliability of structural systems. In these studies, the efficiency of traditional GA techniques was substantially improved by incorporating linkage-learning operators that were used to explore relations among the random variables controlling the safety of a structural system. However, the currently used linkage learning methods were found to be either too complex for easy implementation in routine reliability analyses or too narrowly focused on finding the global optimal solution and thus reduced the capacity of GA to identify local optima. Other techniques especially suitable for exploring relations and linkages among data sets are known as data mining (DM) algorithms. One of the most popular and successful DM tools available in the computer science literature is the a priori algorithm. The purpose of this paper is to propose a hybrid algorithm that would combine the benefits of the pattern identification ability of the a priori DM algorithm to the capability of GA operators to explore new significant search domains. The implementation of the shredding operator and its ability to reduce the computational effort through its self-learning process will further lead to the development of an efficient and robust structural reliability analysis algorithm. This paper will demonstrate that the proposed algorithm will significantly reduce the computational effort associated with determining the probabilistically dominant failure modes of structural systems. Examples are provided to verify the high efficiency and accuracy of the proposed hybrid data mining/genetic shredding algorithm for failure mode exploration, identification, and exploitation.