Akademik Veri Yönetim Sistemi
Journal of Industrial Information Integration, cilt.52, 2026 (SCI-Expanded, Scopus)
The design and development of Digital Twins (DTs) for Cyber–Physical Systems (CPS) entail navigating a vast decision space that encompasses choosing appropriate modeling paradigms, formalisms, computational tools, architectural patterns, deployment frameworks and strategies, and compliance standards. Current DT engineering practices predominantly rely on ad-hoc expert judgment and domain-specific experience, lacking systematic frameworks that integrate structured decision-making processes informed by established domain knowledge and expertise, and supported by advanced reasoning capabilities. This paper introduces TwinCompass, a multi-criteria and hybrid neuro-symbolic decision-making recommendation framework that synergistically combines Multi-Criteria Decision Making (MCDM) methods with Large Language Model (LLM)-based reasoning and symbolic logical inference. TwinCompass leverages a structured and hierarchical decision data model that encapsulates expert DT domain knowledge, represented as decision alternatives, each characterized by a well-defined set of features (criteria). To fundamentally resolve the cognitive bottleneck of manual criteria weighting in MCDM, TwinCompass provides both manual expert-driven and automated LLM-based judgment mechanisms, enabling scalable, adaptive, and context-aware recommendation generation. Furthermore, the framework incorporates a deterministic logical reasoning layer that strictly validates configurations against dependency constraints, compatibility, and prerequisites, ensuring coherent and structurally consistent DT designs. A comprehensive evaluation, including performance analysis of parsing data models, consistency validation of judgments, execution-time correlation analysis, and a qualitative assessment by 20 domain specialists, demonstrates the framework's effectiveness. The empirical results confirm that TwinCompass provides robust decision support, achieving high data-readiness scores under strict filtering conditions. These findings establish the reliability of its recommendations and its suitability for industrial-scale engineering. By addressing this critical gap in the complexity of designing and deploying DTs throughout their lifecycles, this research provides a principled, transparent, and automated decision-making approach that integrates expert knowledge with computational intelligence to support informed, advisable engineering choices.