Development of a shredding genetic algorithm for structural reliability

Abstract Classical genetic algorithms (GA) have been previously proposed for structural reliability analysis but have not been widely adopted because of their inefficiencies. To improve the efficiency of the search process, this paper develops a modified algorithm, referred to as the shredding genetic algorithm (SGA). SGA follows the practice adopted in modern breeding technology where healthy animals are cultivated by interfering with the natural selection process and filtering out pubs with undesirable characteristics using the principle of elitism. By simulating this filtration process, SGA focuses the search around the most important genes thereby improving GA’s efficiency. Following the process used during the human genome project, the filtration criteria are established after shredding each chromosome into strings of genes each of which is associated with a fitness factor. The strings’ fitness factors are assembled into a fitness index matrix that is updated generation by generation as more information about the fitness of chromosomes is gathered. The chromosomes created during the crossover steps are filtered to eliminate those containing strings that do not satisfy a probabilistic filtration standard. Similarly to classical reliability analysis techniques, SGA identifies dominant structural failure modes and also gives detailed information about which random variables are primary contributors to the formation of these failure modes. Furthermore, SGA provides information about linkages that may exist between the random variables that control the safety of structural systems. These linkages allow the identification of sub-mechanisms or partial failures. Such useful information would eventually lead to better control of the safety of structural systems and improve the reliability of designs. Examples are provided to demonstrate the application of the proposed SGA method and its efficiency.