Knowledge-Graph-Governed AI for Engineering Design: A Case Study in S-Lay Inline Structures

Abstract

Engineering organizations store reusable design knowledge in drawings, reports, finite-element studies, spreadsheets, review comments, and expert memory. Much of it is difficult to reuse because component identity, assembly topology, load case, assumptions, response location, and design reasoning are not represented as connected records. This paper presents a case study in applying state-of-the-art AI methods to industrial structural design, using subsea inline structures installed by S-lay as the engineering domain. The work is positioned as a continuation of two earlier S-lay inline-structure studies by the authors: those studies generated the engineering evidence base, and the present paper examines how that evidence can be converted into an ontology-grounded, reviewable AI4D workflow for professional engineers. The proposed method is described independent of any particular software repository. It separates component definitions, assembly rules, behavior evidence, current problem representation, deterministic tools, and governed feedback into EDES, EDAS, EDIKB, EDPR, toolchain, and KEL layers. A large language model is used as a constrained language interface; deterministic tools validate, retrieve, build layouts, plot, and report. The paper explains the S-lay inline-structure ontology, FBS-OAM assembly framing, P-map/APF problem representation, hybrid RAG pipeline, and KEL feedback loop. It does not claim final design sizing, code compliance, validated optimization, or replacement of project-specific analysis.