Abstract In engineering design, standards are instrumental in providing technical definitions and guidelines to designers, manufacturers, and users, reflecting best practices that are widely recognized. These standards, numbering in the hundreds of thousands, encapsulate the current state of technological expertise and aim to promote safety, reliability, productivity, and efficiency in both component and system design. Despite the clear value of standards, the vast quantity and diversity of these documents present significant challenges for designers in retrieving and implementing the appropriate standards for their projects. Moreover, the significant amount of time design engineers exhaust navigating through technical documentation highlights the necessity for more effective methods of engaging with the textual knowledge base represented by engineering standards, as the current manual process is inefficient and time-intensive. This presents an even greater challenge for small and medium-sized enterprises (SMEs) lacking the requisite personnel to adeptly navigate evolving standards. To bridge this gap, this research develops a fundamental understanding of language embeddings and Large Language Models (LLMs) to assist in navigating engineering standards. This work examines the efficacy of utilizing LLMs, such as Bidirectional Encoder Representations from Transformers (BERT), to navigate the extensive corpus of standards and requirements. It also examines three semantic-based approaches for engineering standard retrieval and further validates these approaches through an industry case study. Additionally, the Standards Augmented Requirements Generator (SARG) tool is presented, which provides a domain-specific approach to leveraging Retrieval Augmented Generation (RAG) in design contexts.
Bowen et al. (Fri,) studied this question.