Designing complex space missions are complicated and required various disciplines. This results in a lot of moving parts that must be taken into account for such as development schedule, risks, costs, conflicting metrics, etc. Many tools are developed to assist in the development of space technology. However, as the size of the space mission increases, it is is very complicated to manage space missions. Therefore in this paper, machine learning agents are used to assist System-of-Systems analytic tools and discuss challenges of this approach and future steps.
For space missions, given the extensive design space and a large number of SoS architectures, it becomes extremely hard to employ engineering judgment alone to identify the common features of different well-performing architectures and therefore to take decisions well-supported by quantitative analysis. In the past, the SoS Analytic Workbench developed by the University of Purdue is used to analyze space mission designs. However, with the growing size of mission complexity, the AWB tools is still a bottleneck in the process. To address this problem and support the data mining requirement, the paper proposes the implementation and utilization of Artificial Intelligence (AI) agents to power a specialized space architecture database. These agents can be trained on case studies and then can navigate repositories of documents and publications to find the necessary data to run SoS analysis with tools in the AWB.
Artificial Intelligence Agents to Support Data Ming for SoS Modeling of Space Systems Design
By: University of Purdue
Designing complex space missions are complicated and required various disciplines. This results in a lot moving parts which must be taken into account for such as development schedule, risks, costs, conflicting metrics, etc. Many tools are developed to assist in the development of space technology. However, as the size of the space mission increases, it is is very complicated to manage space missions. Therefore in this paper, machine learning agents are used to assist System-of-Systems analytic tools and discuss challenges of this approach and future steps.
For space missions, given the extensive design space and a large number of SoS architectures, it becomes extremely hard to employ engineering judgment alone to identify the common features of different well-performing architectures and therefore to take decisions well-supported by quantitative analysis. In the past, SoS Analytic Workbench developed by University of Purdue is used to analyze space mission designs. However, with the growing size of mission complexity, the AWB tools is still a bottleneck in the process. To address this problem and support the data mining requirement, the paper proposes the implementation and utilization of Artificial Intelligence (AI) agents to power a specialized space architecture database. These agents can be trained on case studies and then can navigate repositories of documents and publications to find the necessary data to run SoS analysis with tools in the AWB.
In most cases, SoS(System-of-System) cannot always be analyzed with conventional Systems Engineering methodology. Research in the SoS field at Purdue University addressed multiple aspects of SoS, in particular, the dynamic behaviour due to the interactions among constituent systems. As a result, Analytic Work Bench (AWB) was developed within research projects of the Systems Engineering Research Center (SERC) to meet the needs of the US Department of Defense (DoD) for new methodologies to be used for analysis and synthesis of SoS architectures. The AWB is a suite of methods and tools that can be used to achieve a top-level systemic assessment touching different aspects of SoS engineering. The Analytic Work Bench consists of other methods including Systems Developmental Dependency Analysis (SDDA) and Systems Operational Dependency Analysis (SODA). The main function of Systems Operational Dependency Analysis is to addresses the operational domain of a System-of-System, by providing an analysis of the impact of dependencies between constituent systems on the propagation of the effect of disruptions. Systems Developmental Dependency Analysis, on the other hand, looks at the interactions between constituent systems of an SoS for what concerns development and schedule.
The two pictures above show the two methods used in the Analytic Work Bench. It can be seen that one focuses on analyzing how system development relating to scheduling while another one focuses on how dependencies of a system affect the operation of the system.
Traditionally Boolean Search was used to access research in the NASA Technical Reports Server (NTRS). However, this has been problems with this approach. This technique depends heavily on keywords and the accuracy of tagging provided by the author of the articles. This approach leads to results that miss articles that are incorrectly tagged or that containing relevant data but were published against an unrelated primary topic. This is quite big in especially space research, as many of the topics have not been fully discovered/invented yet and hence do not have a known keyword for searches. To solve this issue, a new approach is developed to build a database of space architectures, at the core of which is a new AI technology for topic discovery. These AI topic agents, originally developed by ai-one inc, were used along with document metadata and entity extraction to classify over 60,000 research papers in the NTRS repository. The agents trained for this project were based on the 354 Technology Area Breakdown Structure (TABS) agents from earlier work. Supervised learning was used to train AI agents and it consisted of supplying the agent with keywords, descriptions of each subsystem and samples of the topic from the NASA content, including sentences as well as paragraphs.
The picture above described the Phases of the AI agents training: implementation of the database; development of agents; scoring of the abstracts with agents; presentation of results in Business Intelligence dashboard.
Above is the Spreadsheet showing part of the results of version 5 of the AI agent for Atmosphere Management. The literature sources have been reviewed by Subject Matter Experts, who indicated whether the source is relevant to the topic or not.
In conclusion, the use of Artificial Intelligence to assist System-of-System Analytic Workbench proved to be quite useful. AI agents are able to extract more useful and correct information and research about NASA Gateway Habitat. The tools of the Analytic Work Bench then were able to provide analysis of the impact of dependencies between systems in the operational and in the developmental domain.
Paper: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=9172802