Automated Sensitivity Analysis in Early Space Mission Design

Abstract

Concurrent Engineering in space mission design features several hundred design parameters provided by more than a dozen domain experts. The quality of the design strongly depends on the individual expertise of the people in the team, their experience and collaboration. As part of the design team, a moderator guides the study. In order to bring the study forward, he must have a good overview, ask the design team the right questions and trigger the next actions while keeping the study objectives in mind. The system design evolves quickly; major changes to the baseline are frequent and have to be taken into account. In this dynamic process it is desirable to quickly identify all the effects of a design change throughout the whole system. The paper presents an automated sensitivity analysis approach to evaluate the current design. This analysis can be used directly during the study to identify design drivers and relationships between individual design parameters. The sensitivity data helps to focus on the right components or subsystems and to guide the design process in a well-founded way. The paper explains the implementation of the automated sensitivity analysis algorithm based on a shared system data model, which is capable of holding all the necessary parameters and equations. The implementation in the DLR software Virtual Satellite and the interaction with the user is discussed. A comprehensible example of laying out the fuel tank of a spacecraft shows how the sensitivity analysis can be interpreted and how it can help to improve the design quality. The same example is later used to point out the local relevance of the sensitivity results. In this case, one design parameter is successively changed by three orders of magnitude. Looking at the sensitivity values in each step, the impact of one parameter is completely shifted to another parameter. This effect is the result of the associated dimension of the design parameters. A second implementation calculating dimensionless sensitivity coefficients is then used to create quantifiable results that allow for comparison between the design parameters. In the concluding section on future work we discuss how the automated sensitivity analysis can be used during an actual Concurrent Engineering study with several hundred design parameters.