Extending the Product Architecture Strategies and Effects Database Through Large Language Models

Abstract There are more than 4000 archival publications on the topic of product architecture. Manually synthesizing such a vast body of literature is prohibitive. To ease the burden and bring insight from the literature to product design engineers, we introduced the Product Architecture Strategies and Effects (PASE) database in a previous publication. This database is a collection of relationships between product architecture decisions and their effects as contained in a selection of the literature. Despite the discovery of over 400 relationships, the PASE database was constructed manually from only 17 publications. The present paper describes a way to expand the PASE database, utilizing the data it currently contains to train a set of cascading large language models (LLMs) to rapidly expand the PASE database. In this paper, we carry out the first two steps of the proposed approach and show that it has potential to reduce the effort required to add to the PASE database by orders of magnitude, compared to the manual approach.