| dc.description.abstract | Proposal evaluation of pre-Phase A mission concepts is largely based on the input from subject matter experts who determine the scientific merit of a mission concept based on a number of criteria including: the relevance of the mission objectives to national and international priorities; the existence of a complete set of measurement, instrument, and platform requirements that are traceable to the mission objectives; and several others. The Science Traceability Matrix is a standard tool used to articulate this relevance and traceability and therefore is a key input to this reviewing process. However, inconsistencies in the structure and vocabulary used in the Science Traceability Matrix and other sections of the proposal across organizations make this process challenging and time-consuming. At the same time, as part of the Digital Engineering revolution, NASA and other space organizations are starting to embrace key concepts of model-based systems engineering and understand the value of moving from unstructured text documents to more formal knowledge representations that are amenable to automated data processing. In this line, this thesis leverages transformer models, a recent advance in natural language processing, to demonstrate automatic extraction of science relevance and traceability information from unstructured mission concept proposals. By doing so, this work helps pave the way for future applications of natural language processing to support other systems engineering practices within mission/program development such as automated parsing of design documentation. The proposed tool, called AstroNLP, is evaluated with a case study based on the Astrophysics Decadal Survey. | |
