Comparative Analysis of Global Techniques for Performance and Design Optimization of Launchers

The main goal of this paper is to analyze different methodologies and to quantitatively compare a set of algorithms for multi-objective global optimization, as an initial step toward a multidisciplinary design optimization framework for space transportation systems. Through a comparative analysis based on mathematical benchmarks, hierarchies among several stochastic techniques are proposed. This leads to the identification of two algorithms as the most promising for multidisciplinary design optimization applications: the nondominated sorting genetic algorithm-II (NSGA-II) and an improved version of particle swarm optimization. Then a significant advantage of the latter is highlighted on the problem of trajectory optimization for expendable launch vehicles. In this applicative scenario, the paper also aims at proving the capability of global methods to correctly assess the performance of expendable launch vehicles. Environmental, dynamical, and guidance models for the optimization of ascent trajectories are therefore introduced. Results for seven performance optimization test cases show that global techniques provide comparable payload masses with respect to traditional gradient-based methods. In addition to the trajectory models, historical linear regressions simulating an expendable launch vehicle's design cycle are described. The focus is in this case on the definition of a simple model as a test bench for the verification of the effectiveness of global methods on multidisciplinary design optimization problems, rather than on the quality of the design results. Three widely different expendable launch vehicle design optimization problems are successfully solved by applying the improved particle swarm optimization algorithm. This demonstrates the suitability of the developed global optimization architecture and algorithms for search-space exploration in multidisciplinary design optimization, paving the way for more detailed modeling and more realistic applications.