Optimisation of Categorical Choices in Exploration Mission Concepts of Operations Using Column Generation Method (2412.02521v1)

Abstract: Space missions, particularly complex, large-scale exploration campaigns, can often involve many discrete decisions or events in their concepts of operations. Whilst a variety of methods exist for the optimisation of continuous variables in mission design, the inherent presence of discrete events in mission ConOps disrupts the possibility of using methods that are dependent on having well-defined, continuous mathematical expressions to define the systems. Typically, mission architects will circumvent this problem by solving the system optimisation for every permutation of the categorical decisions if practical, or use metaheuristic solvers if not. However, this can be prohibitively expensive in terms of computation time. Alternatively, categorical decisions in optimisation problems can be expressed using binary variables. If implemented naively, commercially available MILP solvers are still slow to solve such a problem. Problems of this class can be solved more efficiently using column generation methods. Here, restricted problems are created by removing significant numbers of variables. The restricted problem is solved, and the unused variables are priced to test which, if any, could improve the objective of the restricted problem if they were to be added. Column generation methods are problem-specific, and so there is no guaranteed solution to these categorical problems. As such, the following paper proposes guidelines for defining restricted problems representing space exploration mission concepts of operations. The column generation process is described and then applied to two case studies: a ConOps for a crewed Mars mission, in which the design, assembly, and staging of the trans-Martian spacecraft is modelled using discrete decisions; and the payload delivery scheduling of translunar logistics in the context of an extended Artemis surface exploration campaign.