A system engineering recommendation system based on language similarity analysis : an application to space systems conceptual design

Darm, Paul and Berquand, Audrey and Mansilla, Luis and Riccardi, Annalisa; (2022) A system engineering recommendation system based on language similarity analysis : an application to space systems conceptual design. In: 10th International Systems & Concurrent Engineering for Space Applications Conference (SECESA 2022). European Space Agency, Noordwijk. The Netherlands.

[thumbnail of Darm-etal-SECESA-2022-A-system-engineering-recommendation-system-based-on-language-similarity-analysis]
Text. Filename: Darm_etal_SECESA_2022_A_system_engineering_recommendation_system_based_on_language_similarity_analysis.pdf
Accepted Author Manuscript
License: Strathprints license 1.0

Download (1MB)| Preview


In Model-Based System Engineering (MBSE), the creation of complex engineering systems is facilitated by a standard- ised engineering data model and model version control, both of which generate valuable data after each conducted study. However, there are currently few to no approaches, reusing the information and knowledge from previous engineering studies. In this work, we present a new recommendation system, based on a widely adopted engineering data model, defined in the ECSS-E-TM-10-25A technical memorandum. This engineering data model is used by the European Space Agency (ESA), associated partners, as well as in other engineering domains. An engineering model (EM) is a hierar- chical decomposition of an engineering system, providing information about the overall system, design options but also about low-level components. The novel recommendation system leverages a Knowledge Graph (KG) as a unified frame- work for storing multiple EMs. State-of-the-art semantic similarity Natural Language Processing (NLP) techniques are then used to define similarity between higher level information, so called metadata, associated with each EM. Textual information, such as the ”Mission Objectives” of each study, are encoded with a neural language model into a vector representation, which allows to calculate a similarity metric between them, and then compare past-mission metadata with proposed metadata of a new study. In addition, a similarity between lower-level engineering components in the KG is described through the Jaccard metric, which compares components based by the set of parameters that each of them are associated with. By firstly clustering similar engineering designs through their associated metadata and then identifying analogous components in each cluster, the algorithm is able to recommend engineering components for new studies. In the results, the functionality of the approach is demonstrated as a pilot study for spacecraft conceptual design.