A comparative study of Genetic Algorithm and Orthogonal Steepest Descent for aircraft multidisciplinary optimization
Daniel P. Raymer, William Crossley · 2002
This paper reports on a study using a unique form of a Genetic Algorithm (GA) multivariable / multidisciplinary optimization method as applied to aircraft conceptual optimization. A direct comparison was made of this GA method versus an Orthogonal Steepest Descent (OSD) approach, using the same sample aircraft and the exact same analysis methods and executable code. As a sample design, a notional lightweight supersonic fighter was evaluated using both optimizers, and results were compared and contrasted. Note that there are many forms of the basic GA method, and the results obtained from this particular form may not be generically applicable to all forms of GA. Several interesting results emerged. First, it was remarkably easy to incorporate a GA routine with the existing aircraft analysis code. Adding additional variables for optimization proved simple with GA, whereas it was quite difficult with the existing OSD method. Second, the number of designs that needed to be evaluated by the GA exceeded the number needed for the OSD searching method, but not by a large amount. Interestingly enough, even with a large number of generations the GA never quite found an optimum as good as that found by the OSD, but the best GA design was close to that found by the OSD. The differences in execution and final answer were small enough that the advantages of GA described below seem worth pursuing. However, the study also indicated the overall merits of the OSD approach despite its simplicity, it has proven robust and yields a better optimum in fewer computations than GA. Further study will attempt to determine the relative merits of these approaches as additional complexity is added to the optimization process. This research is a company-funded initiative of Conceptual Research Corporation and is being done in cooperation with the Swedish Royal Institute of Technology (KTH) with the assistance and advice of Dr. W. Crossley of Purdue University. Feedback, comments, and suggestions are sincerely requested.