Towards Intelligent Architecting of Aerospace System-of-Systems

System-of-Systems (SoS) are composed of large scale independent and complex heterogeneous systems which collaborate to create capabilities not achievable by a single system, for example air transportation system, satellite constellations, and space exploration architectures. Much of the research effort in the field of SoS has focused on the analysis of these complex entities, while there are still major gaps in developing tools for automated synthesis and engineering of SoS that consider all the various aspects in this problem domain. The gap we address in this paper is a mapping of clusters of SoS architecture alternatives, segmented by performance along multiple metrics, to architectural features. Building upon our previous research where we used a SoS Analytic Work Bench in combination with Model-Based Systems Engineering artifacts to perform analysis of aerospace systems, we propose to build a process for intelligent architecting of aerospace SoS. This process discovers and employs pertinent features in a complex design space to effectively meet the user needs, elevating SoS engineering from retrospective architectural analysis to automated synthesis of new architectures. As a first step towards intelligent architecture of aerospace SoS, we propose to utilize Machine Learning techniques to automate the synthesis phase of SoS. Our hypothesis is that a set of holistic metrics of aerospace architectures (cost, performance, robustness, operational risk, average delay, etc.) can be used to characterize a measure of goodness of architectures, with good architectures on a Pareto front of the multi-dimensional space of holistic metrics of interest. Each architecture or cluster may be then mapped to a set of architectural features, with the goal of identifying which features belong to good architectures. Specifically, we propose to utilize non-parametric regression on a set of training architectures (for example, Neural Networks can deal with mixed real and integer variables) to associate each one with a pattern of features. This mapping will allow the automated process to predict what metrics will be expected from SoS architectures with specific features, and therefore to automatically synthesize architectures that exhibit desired characteristics of goodness. For example, for constellations of satellites, a group of good architecture might have medium cost, high resilience, medium robustness, and low risk, and the architectural features to be mapped to each group can include number of satellites, number of components, type of orbit, type of power system, etc. Since the environment constantly evolves, architectures must adapt, and stochastic optimization can be used to switch between architectures with minimal effort. In this work we illustrate the new version of our aerospace SoS analysis and synthesis framework, which includes Machine Learning techniques to support synthesis of SoS architectures. We demonstrate the application of this process on satellite constellations and discuss challenges of this approach and future steps.