A Modeling Framework for Applying Discrete Optimization to System Architecture Selection and Application to In -Situ Resource Utilization

This paper presents an approach to apply optimization to the selection of a system architecture. Early-stage design decisions that define the subsystem and component technologies that comprise a system architecture are often made with information generated from a limited number of trade studies and a historical background. These decisions can lock in a design too early, resulting in a sub-optimal or even infeasible design. Decisions for systems that have little or no historical background, such as lunar in-situ resource utilization (ISRU) systems, have only analyses and tests to be based on. To enable a wider search of the design space, a modeling framework is discussed that decomposes a system in a hierarchical fashion to describe primary subsystem functions and their technology implementation alternatives. This framework allows a system model to be built that captures all of the subsystem technology alternatives in one reconfigurable model for which the categorical, discrete technology decisions can be treated as design variables in an optimization routine. The NASA ISRU System Model follows this framework and is used as an application of system architecture optimization. A genetic algorithm is used to explore both the discrete and continuous design space of the model. Because the current ISRU System Model has a small set of discrete variables, the performance and computational cost of the genetic algorithm search are compared to a previously developed ISRU optimization scheme that involves full enumeration of the discrete variables followed by gradient-based optimization with the continuous variables. The initial tests of the genetic algorithm approach provide comparable results to the previously used approach, requiring 5100 function evaluations compared to a range of 4800-10,000 function evaluations with the enumeration methods.