An Ontology Construction Framework for Asset Lifecycle Management: Agentic Travelling Approach with Field Application in Shale Oil Wellbore Maintenance

Abstract

Abstract The rapid deployment of AI-native applications in complex wellbore operations requires unified, high-quality data integration to sustain continuous model evolution and reliable asset lifecycle management. However, heterogeneity in standards, semantics, and system architectures across legacy operational datasets limits their immediate applicability for dynamic and agile AI workflows. In alignment with the digital transformation priorities on information management, knowledge integration, and emerging technology implementations, this study develops a Human- and Entity-Centric Agentic Travelling Approach (HATA) for constructing dynamic-static ontology that support adaptive, cross-scenario data integration. This supports the energy transformation by enabling adaptive data integration for complex decision scenarios, validated through 3 typical types of downhole failure cases. HATA introduces a self-organising ontology construction framework composed of three specialised AI agents configured to address data quality challenges in knowledge modelling, and semantic enabled data integration. The planner interprets engineer-determined themes and decomposes them into ontology construction tasks; the traveller explores latent representational spaces to generate ontology candidates optimised for scenario relevance; and the supervisor assesses structural and semantic consistency, providing iterative feedback to guide refinement. The approach was evaluated using operational datasets (2021-2024) from 45 shale-oil wellbores in western China, focusing on wax deposition, scaling, and pump leakage, with corresponding financial management. More than 40 multi-modal parameters including deviation surveys, dynamometer cards, electrical parameters, inspection logs, operational and maintenance (O&M) alarms, and workover records were mapped through the HATA semantic and compared with expert curated benchmarks using a similarity threshold of 0.8 to quantify semantic alignment and structural completeness. Across all evaluated scenarios, the agentic ontology builder demonstrated substantial improvements in interpretability, operational efficiency, and data assembly performance. User intent comprehension achieved expect accuracy, with a single misinterpretation observed. Task planning precision approached optimal levels, while ontology construction cycles were completed within 40 seconds to 2 minutes, requiring minimal supervisory adjustments. Generated datasets reached an average semantic similarity exceeding 0.82 compared with field engineer benchmarks. Importantly, the resulting dynamic ontologies reduced decision-making time for failure diagnosis from approximately 1.5 hours to ≤10 minutes per case. These outcomes demonstrate the potential of HATA to benefit asset lifecycle management and strengthen AI-enabled operational workflows envisioned for wellbore intelligent transformation. This work introduces an innovative agentic travelling methodology for ontology construction, representing a novel innovation of genAI agentic system for ontology and knowledge modelling within shale oil wellbore downhole operations. Unlike static ontology, HATA also supports continuous evolution through digital-twin-driven updates, enabling the generation of adaptive "data blueprints" for predictive analytics. By bridging fragmented legacy datasets with emergent AI workflows, the methodology advances data science priorities in new technology deployment, valuation, and knowledge management, offering a scalable pathway for future energy transformation capabilities.