Asteroid Rescue Mission

This paper is in response to a request for papers from the Lunar and Planetary Institute in Houston, Texas as part of a National University Competition. A human rescue mission to the asteroid 16 Psyche was designed based around a failed Mars mission scenario. The scenario assumed the second human Mars mission, based on the Mars Design Reference Mission 3.0, failed to propulsively capture into Mars orbit, resulting in a higher energy trajectory headed towards the asteroid belt on an intercept trajectory with 16 Psyche. The task was to design a mission that could rescue the astronauts using existing Mars mission hardware prior to the failure of their life support system. Analysis tools were created in the following six disciplines for the design of the mission: trajectory, propulsion, habitat and life support, space rescue vehicle and earth reentry vehicle, space transfer vehicle, and operations. The disciplinary analysis tools were integrated into a computational framework in order to aid the design process. The problem was solved using a traditional fixed-point iteration method with user controlled design variables. Additionally, two other methods of optimization were implemented: design of experiments and collaborative optimization. These were looked at in order to evaluate their ease of implementation and use at solving a complex, multidisciplinary problem. The design of experiments methodology was used to create a central composite design array and a non-linear response surface equation. The response surface equation allows rapid system level optimization. Collaborative optimization is a true multidisciplinary optimization technique which benefits from disciplinary level optimization in conjunction with system level optimization. By reformatting the objective functions of the disciplinary optimizers, collaborative optimization guides the discipline optimizers toward the system optimum. The size and complexity of this design led to severe problems for the advanced optimization methods. The design space was non-smooth, multi-modal, and highly non-linear. Gradient based optimizers could not dependably gather gradient information or find their way out of local minima. Response surface methods produced poor results due to the non-quadratic nature of the design space. Therefore, the traditional fixed-point iteration method proved to be the most easily implemented and produced the best results.