| Title |
Performance Comparison of Genetic Algorithm and Trial-and-error Method in the Member Size Optimization |
| Authors |
Lee Seung-Chang ; Shin Dong-Cheol |
| Keywords |
Genetic Algorithms ; Optimization ; Trial and Error Method ; Dome Structure ; Optimum Design System |
| Abstract |
Genetic algorithms(GA) have been introduced to the field of design optimization during 20 years because it can find near global optimum. In spite of extensive studies of GA applied to truss optimization, it is limited to use for engineering practice as an optimization tool. The causes are originated from the following three points. First, it is not enough to show the advantages of GA compared with the traditional trial and error method. Second, it is wondered if GA can give the reliable results at all cases due to the probability concept and random number. Third, how many iterations of re-analysis will be needed to satisfy the stress and displacement constraints. This study is to answer these questions. For this, various simulations are performed with the typical benchmark problem of 10-bar truss as an initial stage of the study. Simulation results are compared to each other in view of not only the speed of GA but also the number of re-analysis. GA show more effective than trial and error method for small sized problem. As it goes to large problem with many members and load combinations, the advantage of GA is decreased. In the 31-bar truss problem considering actual discrete variable, the performance of GA is similar to that of trial-and-error method. Furthermore, for specific dome structure with over 5000 members, trial and error method is more efficient because it can arrive at the near optimum value. In this paper, the optimum design system is developed based on the trial and error method because it can consider practical conditions; that is, many load combinations, discrete design variables by member database and member grouping technique. |