Satellite Component Selection with Mixed Integer Nonlinear Programming

Spacecraft conceptual design tends to be a laborious and manual procedure where engineers iterate on evaluating performance while having a consistent design, i.e., satisfying budgets at the system level: mass, power, link budget, etc. One interesting part of the problem is that although one could start from scratch to custom design all parts of the spacecraft, development efforts can be saved by picking existing components that satisfy the needed functionality. Building a spacecraft almost exclusively out of catalog components - available commercially off-the-shelf - is especially attractive within the paradigm of CubeSats. Restricting the design to components, however, poses its own set of challenges: iterating through combinations of components that satisfy all budgets can be time consuming as one component might balance the mass budget while unbalancing the power budget. Additionally, as component catalogs keep growing, it becomes impossible to enumerate and consider all combinations, and there is a good chance to miss designs that could come at a lower cost, or a higher performance. Here a novel approach is presented applying mixed-integer nonlinear programming (MINLP), a generalized formulation for component selection. The research applies transformations methods of the nonlinear equations commonly encountered in spacecraft design, in combination with state of the art solvers that can use these transformations to very efficiently solve the problem. The optimization is very fast: less than a minute for the case studied, and the catalog can be expanded without an exponential growth in the solver run time. This in turn makes it very cheap to change the assumptions behind the mission and quickly enumerate optimal designs for different mission constraints. The paper presents a simple model for an Earth Observation satellite, with early conceptual design models for the communications and power system, orbital dynamics, momentum management and satellite lifetime. It then presents the general MINLP component selection formulation and transformations that were applied to the nonlinear models for each different discipline. A small catalog of real and imagined hardware options is presented for four components needed: observation payload, battery, solar cell and antenna. Four different missions are evaluated and compared where the resolution is changed from low to mid to high, and the lifetime of the satellite is changed between 3 and 15 years, showcasing the versatility of this novel method.