Improving Genetic Algorithm Efficiency and Reliability in the Design and Optimization of Composite Structures

Abstract

Genetic algorithms (GAs) often require many iterations for convergence. If the cost for the analysis of each laminate is high, then GA optimization becomes infeasible due to the large amount of CPU time required. A genetic algorithm's ability to find optimal laminate designs that have complicated stacking sequence patterns in an efficient manner may be improved if the GA takes greater advantage of all the information generated throughout the search scheme. In a standard GA procedure, an elitist method is typically implemented where the worst laminate from the chil population is replaced with the best laminate from the parent population. Valuable information that may exist in other laminates of the parent population is no longer utilized once the child population has been created. The present paper suggests new multiple elitist and variable elitist schemes where more than just the best laminate from the old population may be preserved in successive generations providing the GA with additional laminate designs with good performance. These additional designs may contain pieces of the stacking sequence pattern that are vital for achieving the optimal laminate and help the GA converge more rapidly. Results generated by utilizing the multiple elitist and variable elitist methods have shown to yield richer final populations and minor improvements in the computational cost, while maintaining a high level of reliability.