Dissimilarity Measures for Clustering Space Mission Architectures

The application of model transformations to the process of design space exploration and multi-objective optimization allows for comprehensive exploration of an architectural trade space. For many applications, such as the design of missions involving multiple spacecraft, the resulting set of Pareto-optimal solution models can be too large to be consumed directly, requiring additional analyses in order to gain meaningful insights. In this paper, we investigate the use of automated clustering techniques for grouping similar solution models, and introduce and study a number of both generic and domain-specific methods for measuring the similarity of the solution models. We report results from applying our approach to the exploration of the design space of a spacecraft-based interferometry array in a lunar orbit. For purposes of evaluation and validation, results from the application to the case study are correlated with the results from a study in which solution models were clustered manually by groups of domain experts. The results show tradeoffs in the granularity and extensibility of applying clustering approaches to spacecraft mission architecture models. Also, what humans consider to be relevant in assessing architectural similarity varies and is often biased by their background and expertise. We conclude that providing the subjects with a range of clustering tools has the potential to strongly enhance the ability to explore the complex design space of multi-spacecraft missions, and gain deep insights into the trade space.