Structural optimisation via genetic algorithms

[摘要] ENGLISH ABSTRACT: The design of steel structures needs to incorporate some optimisation procedure that evolves the initialdesign into a more economicnal design, where thisnal design must still satisfy all the initial designcriteria. A candidate optimisation technique suggested by this research is the genetic algorithm. Thegenetic algorithm (GA) is an optimisation technique that was inspired by evolutionary principles, suchas the survival of thettest (also known as natural selection). The GA operates by generating apopulation of individuals which 'compete' with one another in order to survive, or di erently stated,in order to make it into the next generation. Each individual presents a solution to the problem.Surviving solutions which propagate through to the next generation are typically 'better' or ' tter'than the ones that had died o , hence suggesting a process of optimisation. This process continuesuntil a de ned convergence criteria is met (e.g. speci ed maximum number of generations is reached),where after the best individual in the population serves as the ultimate solution to the problem.This study thoroughly investigates the inner workings that drive the algorithm, after which an algorithmis presented to face the challenges of structural optimisation. This algorithm will be concernedonly with sizing optimisation; geometry, topology and shape optimisation is outside the scope of thisresearch. The objective of this optimising problem will be to minimise the weight of the structure, itis assumed that the weight is inversely propotional to the cost of the structure. The motive behindusing a genetic algorithm in this study is largely due to its ability to handle discrete search spaces;classical search methods are typically limited to some form of gradient search technique for which thesearch space must be continuous. The algorithm is also preferred due to its ability to e ciently searchthrough vast search spaces, which is typically the case for a structural optimisation problem.The genetic algorithm's performance will be examined through the use of bench-marking problems.Benchmarking is done for both planar and space trusses; the 10 - and 25 bar truss problems. Suchproblems are typically analysed with stress and displacement constraints. After the performance ofthe algorithm is validated, the study commences towards solving real life practical problems. Therststep towards solving such problems would be to investigate the 160 bar truss benchmarking problem.This problem will be slightly adapted by applying South African design standards to the design, SANS(2005). This approach is more realistic, when compared to simply specifying stress and displacementconstraints due to the fact that an element cannot simply be assigned the same stress constraint fortension and compression; slenderness and buckling e ects need to be taken into account. For this case,the search space will no longer simply be some sample search space, but will consist of real sectionstaken from the Southern African Steel Construction Handbook, SAISC (2008). Finally, the researchwill investigate what is needed to optimise a proper real life structure, the Eskom Self-SupportingSuspension 518H Tower. It will address a wide variety of topics, such as modelling the structureas realistically as possible, to investigating key aspects that might make the problem di erent fromstandard benchmarking problems and what kind of steps can be taken to over-come possible issuesand errors.The algorithm runs in parallel with anite element method program, provided by Dr G.C. vanRooyen, which analyses the solutions obtained from the algorithm and ensures structural feasibility.

[发布机构] Stellenbosch University