The dataset comprises approximately 60 sources spanning patents, conference papers, and journal articles published between 1999 and 2026, with a concentration of activity from 2017 onward. Patent landscape analysis via PatSnap reveals that Siemens Aktiengesellschaft is the dominant patent assignee, appearing in at least seven distinct filings across WO, US, EP, CN, and KR jurisdictions — all leveraging knowledge graphs as the core data structure for AI-assisted engineering.

Schneider Electric, Adobe, Georgia Tech Research Corporation, IBM, and JPMorgan Chase also appear as notable filers. On the academic side, institutions including MIT, Carnegie Mellon University, Zhejiang University, Beihang University, and Fraunhofer IPA contribute the bulk of peer-reviewed literature on design rationale capture, semantic networks, and explainable AI in engineering contexts.

The dominant technical approaches cluster around four themes: (1) ontology-driven project knowledge graphs that encode hardware and component relationships; (2) design rationale knowledge networks that capture the "why" behind decisions; (3) explainable AI methods that use knowledge graphs to reconcile black-box model inferences; and (4) traceability link construction between requirements, design artifacts, and source implementations. Together, these themes form a coherent picture of knowledge graphs as the primary infrastructure for rendering AI-generated engineering decisions auditable, reusable, and explainable. WIPO patent data corroborates the global IP footprint of this emerging field.

A key challenge in engineering AI traceability is not just recording what decision was made, but why — the design rationale. Research from the Université de Montréal's Kantara pipeline (2022) proposes using machine learning and NLP to produce a knowledge graph representation of decisions and their rationales, explicitly incorporating historical evolution and traceability. The paper identifies the absence of end-to-end rationale extraction as the key gap in prior work.

The Beihang University study (2023) proposes automatic extraction of design rationale fragments from patent documents and journal articles to construct an interconnected knowledge network. The paper argues that fragmentation of design rationale across documentation is the primary barrier to traceability and reuse — and that a structured knowledge network directly addresses this by enabling designers to navigate from a current design decision back to its documented precedents. Life sciences IP teams at PatSnap apply similar rationale-tracing approaches for regulatory compliance.

Zhejiang University's C-RFBS Model (2021) integrates cognitive process theory with the Requirement-Function-Behavior-Structure model, using knowledge graphs to improve productivity of knowledge records creation and exploration. The model encodes the rationale of deliberation and the context of decision-making directly within the graph structure, making each node's lineage interpretable — essential for auditing AI-generated design choices. IEEE publications have further validated this approach across multiple engineering domains.

Arts et Métiers ParisTech's work (2017) demonstrates that multi-layer graph structures can track the influence of geometric variable modifications across multi-physics design problems, directly addressing the risk of untracked decision propagation in early design stages. The paper explicitly frames traceability as a graph-theoretic problem, anticipating much of the patent literature that followed.