Multiobjective Design Exploration in Space Engineering

Abstract

Most of real world design optimization problems in space engineering are multiobjective design optimization problems that simultaneously involve several competing objectives (Oyama et al., 2002) (Tani et al., 2008) (Oyama et al., 2009) (Oyama et al., 2010). For example, design of a turbopump for liquid rocket engine involves maximization of total head, minimization of input power, minimization of weight, minimization of manufacturing cost, and so on. Another example is trajectory design of a spacecraft where payload weight should be maximized, time required to reach the target point should be minimized, distance from the sun should be maximized (or minimized), and manufacturing cost should be minimized. Many other multiobjective design optimization problems can be easily found, such as reusable space transportation system design, spacecraft design, and Mars airplane design. While a single objective design optimization problem may have a unique optimal solution, multiobjective design optimization problems present a set of compromised solutions, largely known as Pareto-optimal solutions or non-dominated solutions. Each of these solutions is optimal in the sense that no other solutions in the search space are superior to it when all objectives are considered (Fig. 1). Therefore, the goal of multiobjective design optimization problems is to find as many non-dominated solutions as possible to provide useful information of the problem to the designers. Recently, idea of multiobjective design exploration (MODE) (Obayashi et al., 2005) is proposed as a framework to extract essential knowledge of a multiobjective design optimization problem such as trade-off information between contradicting objectives and the effect of each design parameter on the objectives. In the framework of MODE, nondominated solutions are obtained by multiobjective optimization using, for example, a multiobjective evolutionary computation (Deb, 2001), and then design knowledge is extracted by analysing the values of objective functions and design parameters of the obtained non-dominated solutions. There, data mining approaches such as the selforganizing map (SOM) (Kohonen, 1998) and analysis of variance (ANOVA) (Donald, 1998) are used. Recently, MODE framework has been applied to a wide variety of design optimization problems including multidisciplinary design of a regional-jet wing (Chiba et al., 2007a) (Chiba et al., 2007b), aerodynamic design of multi-element airfoil (Kanazaki et al., 2007), and car tire design (Shimoyama, 2009).