Artificial Intelligence Agents to Support Data Mining for SoS Modeling of Space Systems Design

Abstract

The complex and multidisciplinary nature of space systems and mission architectures is especially evident in early stage of design and architecting, where systems stakeholders have to keep into account all the aspects of a project, including alternatives, cost, risk, and schedule and evaluate various potentially conflicting metrics with a high level of uncertainty. Though aerospace engineering is a relatively young discipline, stakeholders in the field can rely on a vast body of knowledge and good practices for space systems design and architecting of space missions. These guidelines have been identified and refined over the years. However, the increase in size and complexity of applications in the aerospace discipline highlighted some gaps in this approach: first, the amount of available information is now very large and originates from multiple sources, often with diverse representations, and useful data for trade space analysis or analysis of all potential alternatives can be easily overlooked; second, the variety and complexity of the systems involved and of the different domains to be kept into account can generate unexpected interactions that cannot be easily identified; third, continuous advancements in the field of aerospace resulted in the development of new approaches and methodologies, for which a common knowledge database is not existing yet, thus requiring substantial effort upfront. To address these gaps and support both decision making in early stage of space systems design and increased automation in extraction of necessary data to feed working groups and analytical methodologies, we propose the training and use of Artificial Intelligence agents. These agents can be trained to recognize not only information coming from standardized representations, for example Model Based Systems Engineering diagrams, but also descriptions of systems and functionalities in plain English. This capability allows each agent to quantify the relevance of publications and documents to the query for which it is trained. At the same time, each agent can recognize potentially useful information in documents which are only loosely connected to the systems or functionalities on which the agent has been trained, and which would possibly be overlooked in a traditional literature review. The search for pertinent sources can be further refined using keywords, that let the user specify more details about the systems or functionality of interest, based on the intended use of the data. In this work we illustrate the use of Artificial Intelligent agents to sort space habitat subsystems into NASA Technology Roadmaps categories and to identify relevant sources of data for these subsystems. We demonstrate how the agents can support the retrieval of complex information required to feed existing System-of-Systems analytic tools and discuss challenges of this approach and future steps.