Showing papers by "William A. Crossley published in 1998"

Empirically-Derived Population Size and Mutation Rate Guidelines for a Genetic Algorithm with Uniform Crossover

Abstract: The Genetic Algorithm (GA) is employed by different users to solve many problems; however, various challenges and issues surround the appropriate form and parameter settings of the GA. One of these issues is the conflict between theory and experiment regarding the crossover operator. Experimental results suggest that the uniform crossover can provide better results for optimization, so many users wish to employ this approach. Unlike for the single-point crossover GA, no established set of guidelines exists to assist in choosing appropriate population sizes and mutation rates when using the uniform crossover. This paper presents the results of an empirical study to determine such guidelines by examining several parameter combinations on four mathematical functions and one engineering design problem. The resulting guidelines appear to be valid over these test problems. They are presented and discussed, with the intent that they may provide assistance to users of GAs with uniform crossover.

Commercial transport aircraft conceptual design using a genetic algorithm based approach

Abstract: Fixed-wing aircraft design is a complex engineering problem, yet the conceptual phase of design is often limited in the number of design variables examined. Further, to begin the design process, many decisions about an aircraft's configuration are based upon qualitative choices of the designer(s). The use of a genetic algorithm (GA) can assist in aircraft conceptual design by reducing the number of qualitative decisions made during the design process while increasing the number of design variables taken into consideration. The genetic algorithm is a search method based on the patterns of natural selection and reproduction common to biological populations. Since the GA operates as a non-calculus based method, discrete and continuous design variables can be handled with equal ease. This paper describes a hybrid approach with the implementation of a GA as a less-biased, automated approach to conceptual aircraft design and the application of CONMIN, a calculus-based method of feasible directions, to refine the results obtained with the GA. Civilian transport class aircraft are the current focus. The resulting optimization-analysis code is used to generate potential conceptual designs for a specified mission. Results from these design efforts are discussed with insight into the use of GAs for conceptual aircraft design.

Multiobjective optimization of a commercial transport aircraft for cost and weight

Abstract: This paper discusses the optimization of a small, medium-range transport aircraft to minimize direct operating cost (DOC), gross weight, and a combination of these two objectives. An effective method for cost optimization of aircraft is desired, especially in the commercial sector where many decisions are costdriven. In the past, many conceptual and preliminary design efforts have assumed a direct relationship between gross weight and cost. However, this relationship is not necessarily the case. The methodology used here for cost prediction incorporates several inputs, not just gross weight. The optimization algorithm CONMIN is used for minimization of the single-objective and multiobjective functions. The design variables are wing loading, thrust-to-weight ratio, aspect ratio, wing sweep, and wing taper ratio. Constraints are imposed on takeoff distance, cruise speed, rate of climb, landing distance, and absolute ceiling. Weighted objectives and a gaming theory are both used as approaches to the multiobjective problem. Results of the optimization studies are presented and discussed.

Using the two-branch tournament genetic algorithm for multiobjective design

Abstract: The two-branch tournament genetic algorithm is presented as an approach to determine a set of Pareto-optimal solutions to multiobjective design problems. Because the genetic algorithm searches using a population of points rather than using a point-to-point search, it is possible to generate a large numher of solutions to multiobjective problems in a single run of the algorithm. The two-branch tournament and its implementation in a genetic algorithm (GA) to provide these solutions are discussed. This approach differs from most traditional methods for GA-based multiobjective design ; it does not require the nondominated ranking approach nor does it require additional fitness manipulations. A multiobjective mathematical benchmark problem and a 10-bar truss problem were solved to illustrate how this approach works for typical multiobjective problems. These problems also allowed comparison to published solutions. The two-branch GA was also applied to a problem combining discrete and continuous variables to illustrate an additional advantage of this approach for multiobjective design problems. Results of all three problems were compared to those of single-objective approaches providing a measure of how closely the Pareto-optimal set is estimated by the two-branch GA. Finally, conclusions were made about the benefits and potential for improvement of this approach.