Browse by author
Lookup NU author(s): Kwabena Adu-Duodu, Yinhao Li, Professor Raj Ranjan, Dr Tejal Shah, Dr Ellis SolaimanORCiD
This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).
The Architecture, Engineering, and Construction (AEC) sector suffers persistent challenges in formalising and automating regulatory compliance, especially for the safe reuse of construction products at their end of life (EoL). Current approaches lack compliance automation aligned with regulation semantics and do not integrate human expertise in the verification of automated decisions. This increases chances of incorrect reasoning over complex regulatory texts. This paper presents the Human-in-the-Loop Semantic Middleware (HiLSeM). Its scientific contribution to Engineering Informatics is a computational formalism for lifecycle-aware regulatory knowledge and verification. It includes a pseudo-automated step (LLM-assisted) to formalise regulatory knowledge into machine-readable structures. It also facilitates traceable, machine-to-machine automated compliance reasoning over isolated data sources while integrating human oversight to review and, where necessary, override automated compliance decisions. The HiLSeM framework builds upon the AEC3PO ontology and related ontology-driven compliance checking streams, extending them with a formal human-verification layer and lifecycle-aware compliance history. Regulatory clauses are analysed using structured RASE (Requirement, Applicability, Selection, Exception) semantics and mapped to interconnected OWL ontologies to deliver executable compliance requirements. The framework was evaluated through staged proof-of-concept case studies covering semantic formalisation, automated reasoning, human verification, and auditable reporting, including a focused raw-versus-corrected formalisation comparison for Clause 5.2 of EN15804. In a clause-level reliability study across selected clauses of varying structure, LLM-assisted extraction achieved precision from 0.50 to 1.00, recall from 0.70 to 1.00, and F1 from 0.67 to 0.93. Gold-standard annotations showed almost-perfect agreement . This work contributes a novel computational formalism that bridges semantic legal knowledge representation, automated reasoning, and auditable human-in-the-loop verification for engineering compliance.
Author(s): Adu-Duodu K, Wilson S, Li Y, Rana O, Wang Y, Ranjan R, Shah T, Solaiman E
Publication type: Article
Publication status: Published
Journal: Advanced Engineering Informatics
Year: 2026
Volume: 76
Issue: Part D
Print publication date: 01/11/2026
Online publication date: 06/08/2026
Acceptance date: 29/07/2026
Date deposited: 10/08/2026
ISSN (print): 1474-0346
ISSN (electronic): 1873-5320
Publisher: Elsevier Ltd
URL: https://doi.org/10.1016/j.aei.2026.105112
DOI: 10.1016/j.aei.2026.105112
Data Access Statement: Anonymised annotators’ prompts (S1), gold-standard annotations (S2), and GPT outputs (S3) used for the GQM evaluation are provided as Supplementary Material and publicly available at Zenodo: https: //doi.org/10.5281/zenodo.17501152.
Altmetrics provided by Altmetric
